DTM 2024 Abstracts

Adenekan Feasibility of Using Smartphone-Based Vibration Perception Assessments for

Monitoring of Sensory Function in Adults at Risk of Developing

Complications from Diabetic Peripheral Neuropathy A1

Agrawal The Role of NGOs in Addressing Non-Communicable Diseases in Rural Odisha: Evidence from an Eye Camp-Based Study A2

Ahmad Glucose Sensing in the Dermis A3

Akturk Persistent, Consistent Clinical Outcomes Using Control-IQ Technology with Dexcom G6 And G7 In an Adult Diabetes Clinic A4

Aloi Treatment of DKA in Patients with Decreased Renal Function A5

Anderson GlucoseReady: Technical Validation of a Novel GXP Platform for Real-Time CGM Capture A6

Anurag Predictive Digital Twin (DT) Of Type 1 Diabetes (T1D) Using Physiology-

Informed Liquid Time Constant (LTC) Neural Networks in Presence of Exercise A7

Ayers The Benefits of Using Continuous Glucose Monitoring to Diagnose Type 1 Diabetes A8

Beltzer Usability Evaluation of the CareSens Air CGM System: Requirements for Regulatory Approval A9

Bem Integration of Continuous Glucose Monitor and Insulin Dosing Data into Electronic Health Records at Diabetes Center (Real World Data) A10

Bem Real-World Outcomes with Early Use of Automated Insulin Delivery Systems in Adults with Type 1 Diabetes A11

Benesch Automated Clamp Testing Procedure for the Clinical Performance Evaluation of Continuous Glucose Monitoring Systems A12

Bergenstal Continuous Glucose Monitoring (CGM)-Based Titration of Once-Weekly Insulin Icodec in Insulin-Naïve Individuals with Type 2 Diabetes

(ONWARDS 9) A13

Bustos Predictive Modeling of Exercise-Induced Glycemic Outcomes Using Generative Deep Learning A14

Calhoun Percent Time Below 54 mg/dL is Strongly Associated with the Number of Weekly Clinically Significant Level 2 Hypoglycemia Events in Type 1 Diabetes (T1D) A15

Cappon Extending the Applicability of ReplayBg Tool for Digital Twinning in Type 1 Diabetes from Single-Meal to Multi-Day Scenarios A16

Cembrowski Negotiating CGM Through the Analytical Minefields of Errant Critical Care Laboratory Glucoses A17

Chattaraj Preclinical Study of a New 10-Day Wear Extended Infusion Set in a Porcine Model of Type 1 Diabetes (T1D) A18

Cossu Leveraging LLM to Improve Efficacy of AGP Reports: A Feasibility Study A19

Cossu Noninvasive Glucose Monitoring: Early Results from the GLUCOMFORT Tattoo Sensor A20

Danaei Nocturnal Hypoglycemia Detection A21

Eichenlaub A Pilot Study of Two Smartwatches with Alleged Glucose Monitoring Functionality A22

Eichenlaub Performance of the 15-Day CareSens Air Continuous Glucose Monitoring System with Optional Calibrations A23

Ejskjaer First Data from Our Fully Digitalized Danish Diabetes PROMS

Questionnaire: Identifying Unmet Psycho-Social Needs in Persons with Diabetes A24

Ejskjaer Full Data Transmission (Clinical and PROMS) from the Patient's Home to the Hospital to the National IT-Infrastructure and Back Again: Individually and Population-Based A25

Faggionato Assessing Beta-Cell Function in Type 2 Diabetes from Continuous Glucose Monitoring Data A26

Faggionato Generating Digital Clones with the UVA/Padova T1D Simulator Identified on CGM and Pump Profiles Via a Bayesian Approach A27

Fields Device Usage in Individuals with Diabetes for Self-Management A28

Flacke Development of a Breath-Based Glucose Estimation Algorithm, A Progress Report A29

Freckmann Current Status of the Working Group on Continuous Glucose Monitoring of the International Federation of Clinical Chemistry and Laboratory Medicine A30

Freckmann Feasibility of a Procedure to Produce Comparator Data for the Standardized Performance Evaluation of CGM Systems A31

García Pérez Needle-Free CGM Based on Magneto-Hydrodynamics A32

Glatzer Including Advanced Glucose Predictions into Diabetes Self-Management: The Accu-Chek® SmartGuide Predict App A33

Goode Exploratory Study of Continuous Glucose Monitoring in the Epidural Space in Swine A34

Haboudal Evaluating Machine Learning Methods for HbA1c Estimation from CGM Data: Impact of Data Pre-Processing on Model Accuracy A35

Herringshaw An Investigation of Cutaneous Reactions to Continuous Glucose Monitors Using a Patient-Centered Survey Tool A36

Herringshaw Cutaneous Reactions to Continuous Glucose Monitors: Predictive Factors of Reaction Severity A37

Holter Continuous Glucose Monitoring with an Osmotic-Pressure Based Continuous Glucose Sensor – Human Pilot Study Results and Next Development Steps A38

Idi Enhancing Safety in Type 1 Diabetes: A Decision Tree Approach to Insulin Pump Malfunction Detection A39

Kannard Feasibility of Multi-Metabolic Continuously Monitoring A40

Kelly Reduced Healthcare Provider Burnout Associated with Use of Spotlight-AQ: Focus on Physical, Mental and Social Wellbeing A41

Kelly Spotlight-AQ Versus Usual Care for Adults with Diabetes: A Trial-Based Complete Case Cost-Effectiveness Analysis A42

Kiehl Real-World Survival of the Medtronic Extended Infusion Set in the United States A43

Klonoff  A44

Kuhlenkötter Downtime Impact Factor Evaluates Downtime in Automated Glucose Clamps A45

Kumbara Evaluating Perplexity and Glucose Level Impact on State-Of-The-Art Generative Pre-Trained Transformer (GPT) Model to Predict Glucose Values at Different Time Intervals A46

Li Identifying Severe Insulin-Deficient Type 2 Diabetes Subphenotype in Electronic Health Records from USA A47

Mader Performance of a Novel Continuous Glucose Monitoring Device in People with Diabetes A48

Mansoor The Effects of Melatonin Supplementation and Continuous Glucose

Monitoring on Glycemic Control, Glucose Variability, And Sleep Quality in a Patient with Type 2 Diabetes: A Case Study A49

Maryniuk Real-Life Use of Text Messaging for Delivery of DSMES for People with Type 2 Diabetes A50

Mastrototaro Novel Wearable Ring with Medical Grade Pulse Oximetry and Other Wellness Metrics A51

Meçani Diabetes Technology and Aviation Medicine: A Brief Review of the Current Evidence and Future Considerations A52

Minoura Towards Minimally Invasive Home-Glycemia Management: Development of a Glycated Albumin Electrochemical Sensor A53

Montaser Evaluating Hyperglycemia Duration and Thresholds as Predictors of A1c Outcomes in Type 1 Diabetes A54

Najafi Smart Offloading for Personalized Diabetic Foot Ulcer Management:

Advancing Remote Patient Monitoring for Comprehensive Risk Evaluation A55

Ogunmuyiwa Hypertension and Diabetes Comorbidity: Factors That Are Associated with Their Joint Occurrence A56

Owen Artificial Intelligence Ready and Equitable Atlas for Diabetes Insights (AIREADI): Purpose and Design A57

Park Implementing a Stiffness Measurement of Red Blood Cells to Predict HbA1c A58

Pellizzari Exploiting Real-World Data and Digital Twins to Develop Effective Formulas for Dosing Insulin Boluses in Type 1 Diabetes Therapy A59

Pleus Accuracy Evaluation of Three CGM Systems Against Different Comparator Devices A60

Prendin Digital Twin for Data Augmentation Enables the Development of Accurate Personalized Deep Glucose Forecasting Algorithms A61

Pryor Accounting for Hypoglycemia Treatments in Continuous Glucose Metrics A62

Rebaud Non-Invasive Glucose Measurement Without Per-Person Calibration Using Raman Spectroscopy A63

Russell Real-World Efficacy of the iLet Bionic Pancreas A64

Schroeder Unlocking Potential: Personalized Lifestyle Therapy for Type 2 Diabetes Through a Predictive Algorithm-Driven Digital Therapeutic A65

Sheehy Integrating Continuous Glucose Monitoring Data into Electronic Health Records: A Solution Conforming to ICoDE 2022 Standards A66

Silva Intradermal Glucose and Lactate Dual Sensor Feasibility for Integrated Diabetes Management A67

Soliman Exploring the Utility and Acceptance of Continuous Glucose Monitor (CGM)

Use Among Hispanic Patients with Insulin-Treated Type 2 Diabetes: A Pilot Focus Group Study A68

Spierling Bagsic Examining Sex Differences in Glycemic Outcomes Among Adults Enrolled in an Inpatient Continuous Glucose Monitoring As Standard of Care (CGM As SOC) Program A69

Stahl Salzman How Sweet it is: Implementing Continuous Glucose Monitoring in Gestational Diabetes A70

Tapsak In Vivo Evaluation of Novel Long-Term Intravascular Continuous Blood Glucose Monitor in a Chronic Ovine Model A71

Thiyagarajah Effectiveness of a Telehealth-Based Digital Therapeutics Program on

Glycemic Control in Type 2 Diabetes Mellitus (T2DM): A Quasi-

Experimental Study A72

Van Name Pharmacokinetic Properties of Once-Weekly Insulin Icodec in Children and Adolescents with Type 2 Diabetes A73

Waki Using Commercial Artificial Intelligence to Assess Nutrients from Photographs of Meals of Type 2 Diabetes Patients A74

Weems Dual Sensing Insulin-Glucose Microneedle A75

Werbaneth Machine-Learning Predictions of Daily Glucose Fluctuations Using Scalp Electroencephalogram (EEG) A76

Yang Quantifying Hypoglycemia Risk Across eHbA1c Levels: Suggesting a Safe Lower Limit for People with Diabetes A77

Yogini C Anti Diabetic Activity of Undaria Pinnatifida and Moringa Oleifera A78

Zehra Point-Of-Care A1c Screening in the Emergency Department A79

Feasibility of Using Smartphone-based Vibration Perception Assessments for Monitoring of Sensory Function in Adults at Risk of Developing Complications from Diabetic Peripheral Neuropathy

Rachel A. G. Adenekan, PhD; Adeyinka E. Adenekan, PhD; Sun H. Kim, MD; Kenneth K. Leung, MD; Srikanth Muppidi, MD; Marilyn Tan, MD; Sandra A. Tsai, MD; Allison M. Okamura, PhD; Cara M. Nunez, PhD; Kyle T. Yoshida, PhD

Stanford University, Department of Mechanical Engineering Stanford, California, USA

adenekan@stanford.edu

Objective:

Although various tools exist for diabetic peripheral neuropathy (DPN) screening, the lack of high-resolution, accessible, fast, and inexpensive standardized methods has led to under-diagnosis, resulting in unnecessary burden on patients and healthcare systems. We aim to assess whether a smartphone-based tactile vibration perception threshold (VPT) measurement platform can be used for early identification of those at risk of developing DPN complications, prior to overt clinical manifestation.

Method:

We assessed the platform’s potential by measuring smartphone-based VPTs (SVPTs) and Rydel-Seiffer tuning forkbased VPTs (TVPTs) in 68 patients with pre-diabetes or diabetes. We used multivariable regression to analyze the relationship between SVPTs and clinical features and used correlation coefficients to assess the relationship between SVPTs and TVPTs.

Result:

SVPTs, measured at the first toe, moderately correlated with TVPTs (R_s = -0.43, p = 0.0019).

SVPTs correlated with clinical markers of DPN incidence (age, HbA1c, and diabetes duration) in adults aged 50 to

69 who have not been previously identified as having impaired vibration perception (F (4, 29) = 4.76, p = 0.00447, Multiple R² = 0.396, Adjusted R² = 0.313, ε = 0.167). SVPTs were positively associated with the interaction term between age and HbA1c (β = 0.118, p = 0.001). SVPTs were negatively associated with diabetes duration (β = 0.098, p = 0.003).

A similar model using clinical markers to predict TVPTs did not fit the data (F (4, 29) = 1.89, p = 0.139, Multiple R² = 0.207, Adjusted R² = 0.0976, ε = 1.921).

Conclusion:

VPTs measured using our accessible platform (< 2 min. test, self-administered, housed entirely within a smartphone) can provide the foundation for early identification and monitoring of DPN.

The Role of NGOs in Addressing NonCommunicable Diseases in Rural Odisha: Evidence from an Eye Camp-Based Study

Manjari Agrawal, Final year medical student; Paresh Bag, SCD Eye Hospital, Optometrist; Dr. Shubhra Gupta, Associate professor of community medicine; Dr. Diksha Singh, Post graduate at community medicine

Pandit Jawaharlal Nehru Memorial Medical College, Raipur Raipur, Chhattisgarh, India

agrawalmanjari13@gmail.com

Objective:

Non-communicable diseases (NCDs) are increasingly prevalent in India, with an earlier onset compared to developed nations. NFHS-5 (2019-2021) reports 17% of men and 14% of women with high blood sugar and 25% of men and 22% of women with hypertension in rural Odisha. Understanding of NCD burden in these remote areas remains limited. This study aimed to evaluate if collaborating with a Non-Government Organization (NGO) could facilitate NCD screening.

Method:

A cross-sectional study was conducted from May 2024 to August 2024 in Titlagarh, Nuapada, and Kalahandi districts of Odisha. Participants attending an eye camp organised by the NGO ‘Services Center for Disabled (SCD)’ were surveyed for demographic, lifestyle, and medical history. Blood pressure (BP) and capillary random blood glucose (RBG) were measured. Hypertension was defined as BP >130-139/80-89 mm Hg, pre-diabetes as RBG 140199 mg/dl, and diabetes as RBG >200 mg/dl.

Result:

Out of 50 participants with blurry vision, 100% agreed to complete the survey. Mean age was 62±12.6 years; 60% were male. Employment rate was 46%, all as daily wagers. Educational level was below high school for all. Tobacco use was 36% (94% male, p=0.001), alcohol use 34%, and 66% consumed meat/eggs regularly. Overweight prevalence was 6%. Hypertension (BP 130-139/80-89 mm Hg) was found in 42% (52% male, p=0.145) and >140/>90 mm Hg in 28%. Pre-diabetes and diabetes were observed in 28% and 8%, respectively. None had been previously screened and were diagnosed with cataract during eye camp with high RBG (400 mg/dl) and positive urine dipstick were referred for medical management prior to surgery.

Conclusion:

Collaboration with NGOs effectively facilitates NCD screening in remote rural areas. Programs like eye camps can help raise awareness, improve screening and management, and enhance health outcomes in these communities.

Dermal Glucose Sensing: Enhancing Precision in Diabetes Management

Khadije Ahmad, MD; Peter Rule, MBA; Bill Van Antwerp

Laxmi Therapeutic Devices, Goleta, California, USA

bill.vanantwerp@gmail.com

Objective:

We aimed to evaluate the accuracy of a novel Continuous Glucose Monitor (CGM) that we developed sensing in the dermis, compared to an FDA-approved CGM, sensing in deep-subcutaneous tissue.

Method:

10 subjects with type-I DM from a US-based clinical site were enrolled in the study. Subjects wore a novel CGM that was compared to frequent YSI whole-blood samples over three in-clinic sessions spanning 7 days.

Simultaneously, subjects wore a commercial CGM (Libre 3), data was analyzed comparing accuracy (MARD) over various glucose Rates of Change (RoC).

Result:

Data from 10 subjects (5M, 5F, median age 36.3±5.9 y) was analyzed. With 619 CGM-YSI matched pairs; the overall MARD of the novel CGM was 5.8%, 98.55% of the values were within ± 20% or ± 20 mg/dL of the YSI reference, while the overall MARD of the commercial CGM was 9.67%, 89.82% of values were within ± 20% of the YSI reference. ISO 5-5, ISO 10-10, ISO 15-15 and ISO 20-20 for the novel CGM were: 56.38%, 86.59%, 95.48% and 98.71% vs. 35.70%, 64.46%, 85.62% and 94.67% for the comparator. Sub-analysis on various RoC showed that the novel CGM was consistently more accurate by at least 4%.

The results suggest that dermal sensing can lead to improved accuracy particularly at high RoC, compared to sensing in deep-subcutaneous tissue.

Conclusion:

Accurate glucose sensing particularly at high RoC is crucial, especially in youth. Preliminary results (n=10) showed that our novel CGM sensing in the dermis with minimal physiological time lag can be more accurate particularly during high RoC.

Persistent, Consistent Clinical Outcomes using Control-IQ Technology with Dexcom G6 and G7 in an Adult Diabetes Clinic

Halis K. Akturk, Andy K. Johnson, Miranda Polin, Edwin D'Souza, Tomas Walker, Jessica Castle, Jordan Pinsker, Laurel H. Messer

Barbara Davis Center for Diabetes, Aurora, Colorado

halis.akturk@cuanschutz.edu

Objective:

The Tandem t:slim X2 and Tandem Mobi insulin pumps with Control-IQ™ technology (Tandem Diabetes Care) recently added support for the Dexcom G7 continuous glucose monitor (Dexcom Inc), allowing users to choose their CGM of preference.

Methods:

At the Barbara Davis Center Adult Clinic, 463 adults (mean age 36±15 years, 59% female, duration of Control-IQ use 4±2 years) started using Dexcom G7 with Control-IQ between November 2023 through June 2024 after previously using Dexcom G6 with their Tandem pump.

Results:

When comparing the 30 days prior to G7 transition with the 30 days after, there were no clinically significant differences in time in range metrics. Median time of sensor use and median time in closed loop were very high (>98% and >94% respectively) when using both G6 and G7 with Control-IQ technology in this real-world sample of one clinical site. Median time in range 70-180 mg/dL was 70.7% with G6 and 71.1% with G7, meeting international consensus guidelines for time in range.

Conclusion:

Dexcom G6 and G7 use with Control-IQ technology showed persistent, consistent clinical outcomes, and allowed for additional user choice for system configuration.

Treatment of DKA in Patients with Decreased Renal Function

Joseph A. Aloi, MD; Anderson Schrader, ME; Paul D. Chidester, MD

Atrium Wake Forest Baptist Medical Center, Winston-Salem, NC, USA

jaloi@wakehealth.edu

Objective:

To evaluate the impact of renal insufficiency on hypoglycemia rates in patients treated for Diabetic Ketoacidosis (DKA) utilizing a computerized insulin dosing algorithm

Method:

The incidence of hypoglycemia and duration of insulin infusion in patients with DKA was examined by the degree of renal insufficiency. An analysis of the de-identified EndoTool IV (ETIV) database was undertaken. ETIV is an insulin dosing software that adjusts insulin dosing based on the degree of renal impairment and an estimate of residual active insulin. Estimated glomerular filtration rate (eGFR) is calculated by ETIV utilizing the formula by Levey (Ann Intern Med. 2009 May 5; 150(9): 604–612). Patients were stratified into groups by level of eGFR:

greater than 30, 20-30, 10-20 and less than 10 ml/min.

Result:

For the year 2023, ETIV database identified 8059 patients treated for DKA.1491 patients had an eGFR < 30 ml/min. A total of 190,289 blood glucose measurements were obtained. The incidence of severe hypoglycemia, BG < 40 mg/dl as a percentage of all glucose values obtained was less than .01% for each group. There was an increase in the incidence of hypoglycemia, BG <70mg/dl, (0.25% vs .5%, p<.0001) for all eGFR groups < 30 compared to those patients with an eGFR > 30 ml/min. There was no difference in the incidence of hypoglycemia between the groups based on treatment goal range. Time on infusion was longer (24.5 vs 29 hrs, p < .001) in those patients with an eGFR 10-30compared to those with an eGFR > 30ml/min.

Conclusion:

ETIV appropriately manages DKA patients with renal insufficiency. Those patients with the most severe impairment in renal function had no increase in severe hypoglycemia events.

GlucoseReady: Technical Validation of a Novel GxP Platform for Real-Time CGM Capture

David Anderson, PhD; Steve Polyak, PhD; Michael Merickel, MS; Brian Severson, BA; Allen Best, JD; Jonathan Lieu, None; Christian Djurhuus, MD, PhD; Jonathan Goldman, MD

Clinical ink, Winston-Salem, North Carolina, USA

david.anderson@clinicalink.com

Objective:

FDA recommends continuous glucose monitoring (CGM) for diabetes clinical trials. Real-time monitoring demands high-speed data frameworks. Here, we sought to develop a fully GxP-compliant platform that integrates with CGM devices and rapidly transmits regulatory submission quality data to study servers.

Method:

Software acquired glucose readings directly from Dexcom G6 CGM transmitters and sensors via Bluetooth. Live transmitter samples were considered real-time readings; backfill samples were excluded from real-time classification. Software synchronized with study servers after each sample acquisition, pending network connectivity. Descriptive statistics were evaluated.

Result:

CGM readings (n=5,996) were generated over a 24 day study period. Median inter-sample interval was 5.0 minutes, and 99.4% of all sample intervals were between 4.9 and 5.1 minutes. Sample validity was high (98.6%), where 1.3% of samples were lost to sensor warmup and 0.05% of samples were lost to sensor error. Real-time measurements made up 96.4% of readings. Median transmitter sync time was 0.50 seconds (IQR: 0.26 – 0.75 seconds). Server upload times were fastest at 12 seconds; median upload time was 10.3 minutes (IQR: 5.3 – 25.3 minutes). Participant showed normal glycemic variability (19.7%), time-in-range (97.3%), and glycemic risk index (3.6%), with few Level-1 (n=3) and Level-2 (n=1) hypoglycemic events.

Conclusion:

We developed a fully GxP-compliant high-speed data platform directly integrated with CGM devices. There is potential for low latency data availability – on the order of seconds – and further investigation is required to understand sources of server upload lag, for example network connectivity. Future development of this platform will include integrated hypo forms triggered by CGM data streams. Further work is needed to understand network dynamics of this framework in a larger study cohort.

Predictive Digital Twin (DT) of Type 1 Diabetes (T1D) Using Physiology-Informed Liquid Time Constant (LTC) Neural Networks in Presence of Exercise

Anurag Anurag; Ayan Banerjee, PhD; and Sandeep Gupta, PhD; Marzia Cescon, PhD

University of Houston, Houston, TX, United States 77004

ch7210026@iitd.ac.in

Objective:

In this work, we propose a novel strategy to train a machine learning (ML) based Digital Twin (DT) of a patient with Type 1 Diabetes (T1D), combining a large-scale metabolic model and a liquid time constant (LTC) neural network architecture. The DT is able to replay various realistic 24-hour-long scenarios and predict the blood glucose (BG) levels in presence of American Diabetes Association (ADA) recommended level of exercise.

Method:

Using an FDA-accepted large-scale metabolic simulator, T1D data was collected for a representative patient from 20 scenarios, each lasting 24 hours. Standardized meals were served at breakfast, lunch, and dinner with carbohydrate content sampled from normal distributions (means: 50, 75, 75 [g]; SDs: 5, 7, 7 [g]). Insulin doses were based on the insulin-to-carbohydrate ratio. Mealtimes were 7 a.m., 12 p.m., and 7 p.m. Each scenario included 45 minutes of aerobic activity at 5:30 p.m. A nominal scenario without randomization was also simulated for testing. After preprocessing, the 20 scenarios dataset was split into 90-10% for training and validation. The LTC neural network was trained for 200 epochs with a batch size of 16. The model's performance was compared with Long Short-Term Memory (LSTM) and Continuous Time Recurrent Neural Networks (CTRNN).

Result:

Evaluating the model performance in testing set resulted in a mean absolute error (MAE) of 4.24, 10.51, 16.58 [mg/dL], and root mean square error (RMSE) of 6.82, 17.43, 23.01 [mg/dL] for LTC, LSTM and CTRNN models, respectively.

Conclusion:

The proposed proof-of-concept LTC-NN DT model of a patient with T1D predicts clinical blood glucose data reliably in both short- and long-term scenarios, outperforming LSTM and CTRNN models.

The Benefits of Using Continuous Glucose Monitoring to Diagnose Type 1 Diabetes

Alessandra T. Ayers, BA; Cindy N. Ho, BA; Jenise C. Wong, MD, PhD; David Kerr, MBChB, DM, FRCPE; Julia K. Mader, MD; David C. Klonoff, MD, FACP, FRCP (Edin), Fellow AIMBE

Diabetes Technology Society, Burlingame, California, USA

ayers@diabetestechnology.org

Introduction:

Identification of Stage 2 (S2) Type 1 diabetes (T1D) can allow prevention therapy to be prescribed to delay progression to Stage 3 (S3) T1D. CGM data might be useful for identifying an optimal time to administer this treatment to avoid serious and costly health complications such as DKA.

Method:

We searched PubMed to identify relevant studies using the search string: [(“Type 1 Diabetes” OR “T1D”) AND ("continuous glucose monitor" OR CGM) AND (“autoantibod*” OR “autoimmunity”)]. We aimed to determine CGM metrics that are correlated with future development of S3 T1D.

Results:

After screening 25 studies, we identified eight studies where CGM metrics were analyzed to predict progression to S3 T1D. Seven studies indicated that CGMs may be useful for predicting this progression, and five of those studies indicated that time spent with BG ≥140 mg/dL is a statistically significant predictor of progression to S3 T1D. For instance, Steck et al. (2022) found that progressors spent 21% of time ≥140 mg/dL compared to 3% for nonprogressors (P<0.0001), and concluded that >10% of time with BG >140mg/dL is associated with a high risk of progression to clinical diabetes within one year for Ab+ children. Other CGM metrics differing between progressors and non-progressors in at least one study included maximum daytime blood glucose (BG), mean BG, variation in BG (range, SD, MAGE), and TIR.

Discussion:

Although CGM has been identified as a tool to predict progression to S3 T1D, there is no consensus on which particular statistics and thresholds to use. Future research can likely achieve consensus to use CGM as a tool to determine when to offer treatments to delay the onset of S3 T1D.

Usability Evaluation of the CareSens Air CGM System: Requirements for Regulatory Approval

Anne Beltzer, PhD; Marta Gil Miró, MSc; Stefan Pleus, PhD; Moon Hwan Kim, MSc; Hakhyun Nam, PhD; Guido Freckmann, MD

Institut für Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universität Ulm, Ulm, Germany

anne.beltzer@idt-ulm.de

Objective:

Usability requirements for medical devices seeking CE mark became more stringent with the implementation of the Medical Device Regulation (MDR) 2017/745 in 2021. Based on the risk assessment of a product, usability evaluations aim to detect design flaws and unexpected use errors (formative evaluations) or to assess safe and effective use (summative evaluations). This work shows the summative usability evaluation of the CareSens Air continuous glucose monitoring (CGM) System (i-SENS, Inc., Korea) that resulted in obtaining the CE mark.

Method:

The usability testing of the CareSens Air CGM System and its corresponding mobile application was performed according to the international standard IEC 62366-1. A total of 15 adults with type 1 diabetes operated the medical device under simulated conditions of use. The success rate was calculated based on correct task completion and observed use errors. Furthermore, the participants’ subjective perception of the CGM usability was quantified using the System Usability Scale (SUS) questionnaire.

Result:

Participants handled the device with ease and quickly became familiar with the CGM system. The user manual contained pertinent warnings and safely guided the participants through the tasks. Observed use errors did not cause any harm to the user as assessed by risk analysis or were due to the simulative setting. Participants rated the usability of the system between “excellent” and “good”. SUS score was 74, which is considered acceptable.

Conclusion:

This summative evaluation did not reveal any critical situation in the handling of the device. Therefore, the safe and effective use of the CareSens Air CGM System was confirmed.

Integration of Continuous Glucose Monitor and Insulin Dosing Data into Electronic Health Records at Diabetes Center (Real World Data)

Robert Bem, MD, PhD, MHA; David Vavra; Tomas Neskudla, MD; Martin Haluzik, MD, DSc

Diabetes Center, Institute for Clinical and Experimental Medicine Prague, Czech Republic

robert.bem@ikem.cz

Objective:

Continuous glucose monitor (CGM) and insulin dosing data integration into the electronic health record (EHR) plays a crucial role in the care of people with type 1 diabetes (T1D).The study assessed the feasibility of integrating semi-automated CGM and insulin data into the EHR and their association with diabetes control parameters for adults with T1D at our diabetes center.

Method:

Between 2012 and 2023, data from all adults with T1D treated at our Diabetes Center were used for this analysis (from 1765 in 2012 to 2609 subjects in 2023). We assessed the integration of CGM and insulin data into the EHR, including downloading data from sensors, smart pens, and insulin pumps. Data were integrated semi-automatically by processing PDF records sent via email or downloaded during visits, and we evaluated their association with diabetes control parameters.

Result:

Since 2012, there has been a significant increase in the percentage of individuals with integrated CGM and insulin data in EHR (2012: 5.3%; 2016: 41.2%; 2020: 59.6%; 2023: 78%; all p<0.05). The number of integrated data points per person significantly increased during the observed period. In 2023, 22.9% of individuals had data downloaded 12 times, 46.6% 3-5 times, and more than 5 times 8.5%. A decrease in HbA1c was recorded (2012: 8.5%; 2016: 8.1%; 2020: 8%; 2023: 7.3%; all p<0.05), with significantly lower HbA1c levels observed in individuals with integrated data compared to those without (e.g., 7.3% vs. 7.6% in 2023).

Conclusion:

In our real-world study, integrating CGM and insulin dosing data into the EHR was associated with improved diabetes control in adults with T1D.

Real-World Outcomes with Early Use of Automated Insulin Delivery Systems in Adults with Type 1 Diabetes

Robert Bem, MD, PhD, MHA; David Vavra; Michal Dubsky, MD, PhD, FRSPH; Martin Haluzik, MD, DSc

Diabetes Center, Institute for Clinical and Experimental Medicine Prague, Czech Republic

robert.bem@ikem.cz

Objective:

Automated Insulin Delivery systems (AIDs) improve glucose control, reduce the risk of hypoglycemia, and enhance the quality of life in individuals with Type 1 Diabetes (T1D). This study aims to evaluate changes in glucose control during the first year of AIDs use in a real-world cohort of adults with T1D.

Method:

In 2023, we treated 221 adults with T1D initiating AIDs at our center. The cohort included 45.7% men, with a mean age of 42.7±15.1 years and a diabetes duration of 24.2±11.6 years. AIDs (MiniMed 780G with SmartGuard, Tandem with Control-IQ, or YpsoPump with CamAPS FX) were prescribed to 53 subjects previously on multiple daily injections (MDI) and 168 subjects on insulin pumps (IP) without AIDs. We assessed basic characteristics, glucose control parameters [HbA1c, Time in Range (TIR 3.9-10), Time Below Range (TBR <3.9), Time Above Range (TAR >10), average glucose, coefficient of variability (CV)], and the presence of diabetes complications and comorbidities.

Result:

Significant improvements were observed after AIDs initiation: HbA1c decreased (7.7±3.2 to 7±2.9 %; p<0.001), TIR increased (59.4±16.3 to 73.8±12.5 %; p<0.001), TBR decreased (5.2±5.5 to 2.5±3.1 %; p<0.001), and TAR decreased (35.4±17.2 to 23.7±12.7 %; p<0.001). Average glucose levels (9±1.6 to 8.3±1.1 mmol/l; p<0.001) and CV (36.8±8.4 to 33.3±6.4 %; p<0.001) also improved. Patients previously on MDI showed more significant improvements in TIR, TBR, and CV than those on IP (all p<0.02).

Conclusion:

Initiating AIDs significantly improved diabetes control in adults with T1D, with better results in those initially on MDI.

Automated Clamp Testing Procedure for the Clinical Performance Evaluation of Continuous Glucose Monitoring Systems

Carsten Benesch, PhD; J. Hans DeVries, MD; Mareike Kuhlenkötter, MSc

Profil, Neuss, Germany

carsten.benesch@profil.com

Objective:

Comparing the performance of different Continuous Glucose Monitoring (CGM) systems is challenging due to the lack of comprehensive guidelines for clinical study design. In particular, the absence of concise requirements for the distribution of comparator blood glucose (BG) concentrations and their rate of change (RoC) that are used to evaluate CGM performance impairs comparability.

Method:

In 2024 experts proposed requirements for an adequate distribution of comparator glucose measurements and a testing procedure for the manipulation of glucose levels. The proposed testing procedure consists of a meal to raise BG concentrations and a delayed individualized insulin bolus to reach the normo-/hypoglycemic range without any control of the RoC between both ranges. However, especially rapidly falling BG concentrations deserve special interest because of the risk for serious hypoglycemia and, due to the well-known time lag of the CGMs, there is a potential for large differences between measured CGM values and the actual BG concentration. We therefore programmed our automated clamp device to follow the glucose profile proposed by the experts. All relevant BG regions are covered during a predefined period of time with predefined upward and downward RoCs to compare CGM systems.

Result:

In an in-vitro experiment, we achieved excellent agreement between the BG values and the target levels and time in normo-, hyper- and hypoglycemia (precision 1.0% - 3.3%) as well as highly reproducible positive and negative RoCs (control deviation: -0.3mg/dL, absolute control deviation: 3.7mg/dl).

Conclusion:

This work is a step towards establishing a future standard for the performance evaluation of CGM devices, ensuring that desired periods are spent in the hypo-, normo- and hyperglycemic ranges, with desired RoCs in a safe and reproducible way.

Continuous Glucose Monitoring (CGM)Based Titration of Once-Weekly Insulin Icodec in Insulin-Naïve Individuals with Type 2 Diabetes (ONWARDS 9)

Richard Bergenstal, MD; Björg Ásbjörnsdóttir, MD, PhD; Tanvir Bari, MD, PhD; Satish Hulkund, MSc; Yvonne Winhofer, MD, PhD; Carol Wysham, MD

International Diabetes Center, HealthPartners Institute, Minneapolis, MN, USA

richard.bergenstal@parknicollet.com

Objective:

ONWARDS 9 (NCT05823948) was a 26-week, treat-to-target, single-arm, phase 3b study designed to explore the effect of continuous glucose monitoring (CGM)-based titration of once-weekly insulin icodec (icodec) on glycemic control and safety outcomes in individuals with type 2 diabetes (T2D).

Method:

Insulin-naïve adults (n=51) with T2D (HbA_1c 7.0–11.0%) initiated icodec treatment with 70 units once weekly and thereafter titrated weekly based on CGM values. The primary endpoint was change in HbA_1cfrom week 0 to week 26. CGM metrics were collected throughout the study (weeks −2 to 31) including time in range (TIR; 70–180 mg/dL), time above range (TAR; >180 mg/dL) and time below range (TBR; <54 mg/dL). The number of severe hypoglycemic episodes was also collected through participant reporting.

Result:

At week 26, from a mean of 8.18% at week 0, the estimated change in HbA_1c was −0.99%-points (95% confidence interval: −1.17; −0.81, p <0.0001). From week −2–0 to week 22–26, observed mean TIR increased from 54.36% to 76.38%, mean TAR reduced from 45.16% to 22.90%, while mean TBR was 0.03% at weeks −2–0 and 0.04% at weeks 22–26. No severe hypoglycemic episodes were reported during the study.

Conclusion:

From week 0 to week 26, there was a statistically significant reduction in HbA_1c. A concomitant clinically meaningful increase in TIR and decrease in TAR was observed, while TBR remained very low with no episodes of severe hypoglycemia reported throughout the study. The internationally recommended CGM targets for TIR, TBR_{<54 mg/dL} and TAR_{>180 mg/dL} were achieved. These findings suggest that CGM-based titration of icodec is feasible in adults with T2D.

Funded by Novo Nordisk A/S

Predictive Modeling of Exercise-Induced Glycemic Outcomes Using Generative Deep Learning

Oriol Bustos, BSc.; Júlia Soler, Msc.; Omer Mujahid, PhD; Josep Vehí, PhD

Modeling, Identification, and Control Engineering Laboratory, Insitut d’Informàtica i Aplicacions, Universitat de Girona, Girona, Spain

oriol.bustos@udg.edu

Objective:

Physiological models cannot account for the entire complexity of glucose homeostasis, and often overlook critical dynamics present in real patients. This is especially notorious in simulators including physical activity. Data-driven models, however, learn directly from real-life data, including lifestyle. Instead of predicting glucose profiles, datadriven models can predict glycemic outcomes over a period of time, reproducing the variability observed in real data.

The objective of this work is to develop a state-of-the-art methodology based on a generative deep learning model capable of simulating the effect physical activities have on glycemic outcomes.

Method:

A novel Deep Convolutional, Conditional, Wasserstein Generative Adversarial Network (DCCW-GAN), is used to generate virtual patients. Inputs to the model are: delivered insulin, carbohydrates intake, alongside a derived measure of exercise intensity. Data from 60 standard insulin pump patients performing aerobic exercise were taken from the T1DEXI dataset for training and validation. Change in glucose during exercise and time in range 70-180 mg/dL (TIR) on active days and sedentary days generated by the model were compared against real clinical data.

Result:

The median TIR for generated and real patients on sedentary days was: 77.43% and 77.26% respectively (p = 0.11). On active days, the mean TIR was 79.17% and 81.25% for generated and real patients respectively (p = 0.73). The median change between baseline and nadir glucose levels during exercise was -13.09 mg/dL for generated data and 12.0 mg/dL for real data (p = 0.54).

Conclusion:

Our study confirms the potential of GANs in predicting glycemic outcomes in response to exercise. By realistically replicating known glucose responses, our model demonstrated predictive accuracy and real potential utility in managing exercise-induced glucose variations.

Percent Time Below 54 mg/dL Is Strongly Associated with the Number of Weekly Clinically Significant Level 2 Hypoglycemia Events in Type 1 Diabetes (T1D)

Peter Calhoun; Roy W. Beck; Zoey Li; Susan Peers; Michael C. Riddell; Jennifer L. Sherr; Richard T. Liggins

Jaeb Center for Health Research, Tampa, Florida, USA

pcalhoun@jaeb.org

Objective:

CGM time below range targets (<4% T<70 for Level 1 and <1% T<54 mg/dL for Level 2 [L2]) are used in the clinical management of hypoglycemia in individuals with T1D. For people with T1D and caregivers, clinically significant L2 events (glucose <54mg/dL for >15 consecutive minutes) are also important. This analysis describes the associations between percent T<54 and L2 hypoglycemic episodes in pediatric and adult individuals with T1D using self-monitoring of blood glucose (SMBG), or diabetes-related technologies including real-time CGM and automated insulin delivery systems (AID).

Methods:

Associations between T<54 and L2 hypoglycemia event rates, as measured using CGM (real-time, or for SMBG users interval-use blinded CGM), were examined from 8 trials over a 3-6 month period in participants stratified by age and technology use.

Result:

Of 1530 participants (aged 6-17: N=442 or aged ≥18: N=1088), 14%, 42% and 43% were SMBG, CGM, or AID users, respectively. T<54 was correlated with L2 hypoglycemic event rates across all technology types and age strata (r≥0.87). For participants meeting consensus T<54 target (<1%), the chance of developing ≥1 L2 hypoglycemic episodes/week varied by technology use: 40%, 33% and 22% for SMBG, CGM or AID users respectively. At the T<54 target threshold (1%), the likelihood of having ≥1 vs. ≥2 L2 events/week was >99% vs. 29%, and each 0.1% increase in T<54 multiplied the odds of ≥2 L2 hypoglycemic episodes by 1.6, irrespective of age and technology use.

Conclusion:

Percent T<54 is highly associated with the number of weekly clinically significant L2 hypoglycemic events in adults and adolescents with T1D, irrespective of technology use further strengthening the utility of discussing this metric in clinical care.

Extending the Applicability of ReplayBG Tool for Digital Twinning in Type 1 Diabetes from Single-Meal to Multi-Day Scenarios

Giacomo Cappon, PhD; Simone Del Favero, PhD; Andrea Facchinetti, PhD

University of Padova, Department of Information Engineering (DEI) Padova, Italy

giacomo.cappon@unipd.it

Objective:

Digital twins (DT) for individuals with type 1 diabetes (T1D) are personalized models of their glucoregulatory systems, generated by integrating data from continuous glucose monitoring (CGM) sensors, insulin pumps, and food diaries. DT are invaluable for assessing, developing, and optimizing new personalized treatments for T1D. Recently, an open-source tool, ReplayBG, has been developed to create such DTs. ReplayBG can currently work using data of single-meal scenarios, typically encompassing 8-12 hours of data, and requires initial steady state conditions, which means the data used to create the DT must begin sufficiently after the last meal/insulin event. These constraints limit the usability of data for constructing DTs and utilizing ReplayBG. This work extends the capabilities of ReplayBG to support the creation of DTs using wider portions of data spanning multiple days.

Method:

ReplayBG's model was expanded with i) a “multi-stomach” system to simulate distinct meal absorption dynamics observed throughout a single day, ii) a time-varying profile for insulin sensitivity to account for intraday variability in insulin response, and iii) a model of CGM sensor error to simulate realistic data. The model identification process was refined by replacing the previous Markov Chain Monte Carlo methodology with a more advanced ensemble approach and adapting it to encompass multi-day intervals.

Result:

The impact of these expansions was evaluated using the OhioT1DM dataset, which comprise data from 12 patients with T1D. Results show that, compared to the original version, both model fit and the amount of usable data improved significantly, demonstrating the superiority of the new extended ReplayBG.

Conclusion:

This work enhances ReplayBG's capability to accurately model and simulate real-world scenarios of T1D management increasing its practical applicability in clinical and research settings.

Negotiating CGM Through the Analytical Minefields of Errant Critical Care Laboratory Glucoses

George Cembrowski, MD,PhD; Jialin Qiu; Jordyn Stolee; Yun Huang, PhD

Alberta Precision Laboratories, Edmonton, Alberta, Canada

cembr001@gmail.com

Objective:

In the future, evaluations of hospital-based CGM systems will be attempted in hospital ICUs equipped with paired, automated blood gas/chemistry analyzers. Based on 3 years of thrice daily measurement of fresh whole blood on paired GEM 4000s, a prevalent ICU analyzer, we demonstrate significant periods of inaccurate glucose reporting that would compromise CGM evaluation.

Method:

At Kingston Hospital, from July 2018 to June 2021, after morning, day and evening testing, glucose was retested on one patient specimen by central laboratory, side-by-side GEM 4000s. We used Dahlberg’s analysis of the duplicate glucose values to determine short-term, intermediate and long-term analytical variation of the GEM 4000. As the GEM uses disposable test cartridges, we correlated the variation to the dates and times of the cartridge replacements.

Results

Over three years, 384 cartridges were used, 209 in one GEM and 175 in the other. Variation was measured within 335 double cartridge combinations and across 28 testing periods (9.33 days). The average patient glucose was 8.1 mmol/L. The within-cartridge variations follow: 1.8%(25P), 2.5%(50P), 3.3%(75P), 5.6%(95P) and

8.0%(99P). Over the 28 sequential testing periods associated with use of the same cartridge combinations, the average variation dropped from 2.1% to 0.0%. Within 28-day variations were assessed over 3 years: 2.5%(25P), 2.9%(50P), 3.2%(75P), 4.5%(95P) and 5.6%(99P).

Conclusion:

GEM’s glucose variation is related to using dual analyzers that require frequent cartridge changes. As comparisons between the GEM and CGM would complicate CGM hospital evaluations, we recommend that CGM manufacturers resort to either their usual BGM or the newly FDA-approved Nova Primary which achieves the maximum acceptable long term variation for glucose analyzers engaged in diabetes research, 3.5%.

Preclinical Study of a New 10-Day Wear Extended Infusion Set in a Porcine Model of Type 1 Diabetes (T1D)

Sarnath Chattaraj, PhD; Gina Zhang, PhD; Jessica Ortigoza-Diaz, PhD; Davy Tong, MS; Timothy Kwa, PhD; Ohad Cohen, MD

Medtronic Diabetes, Northridge, California, USA

sarnath.chattaraj@medtronic.com

Objective:

Most of the on-market insulin infusion sets are approved for 2-3 days, with the Medtronic Extended infusion set (EIS) being the only IS approved for up to 7-day use, which doubles the length of time an IS can be worn. The EIS allows users to safely stay on insulin pump therapy with fewer interruptions/insertions, while providing enhanced convenience and comfort in diabetes management. To further extend the wear duration, a new infusion set was developed (Photon Infusion Set, or PIS) and assessed in a type 1 diabetes (T1D) porcine model.

Method:

Four drug-induced T1D Yucatan pigs wore the PIS multiple times on the abdomen, and the duration of wear was compared to that of historical EIS control (N=28). Sets were continuously infused with insulin aspart subcutaneously via a pump for 10 days or until failure. Blood glucose was measured 5 times/day and all pigs wore a Medtronic glucose sensor ≥ 5 cm away from the set. Physical examination and serum chemistry panel tests were conducted for safety evaluation.

Result:

Fifteen PIS wears were completed, with all sets lasting 10 days. No significant safety concern was identified with

PIS wear. There was no significant difference detected in total daily dose (TDD) of insulin between the PIS and the EIS control, and the TDD in the PIS group did not change during the entire wear duration (Day 1: 14.56 ± 0.61 U, Day 9: 14.23 ± 2.39 U).

Conclusion:

This study demonstrated the safety and efficacy of PIS 10-day wear, when infusing insulin aspart in a porcine model of T1D. A clinical study of the device will further demonstrate safety and performance of the PIS.

Leveraging LLM to Improve Efficacy of AGP Reports: A Feasibility Study

Cossu Luca, MSc; Cappon Giacomo, PhD; Annalisa Deodati, MD, PhD; Riccardo Schiaffini, MD, PhD; Stefano Cianfarani, MD and Facchinetti Andrea, PhD

Department of Information Engineering, University of Padova Padova, Italy

cossuluca@dei.unipd.it

Objective:

The Ambulatory Glucose Profile (AGP) report is essential for clinicians to assess and improve glucose control in individuals with Type 1 Diabetes (T1D). Understanding the day-to-day fluctuations of glucose levels and identifying patterns over several days are crucial elements of AGP analysis and the numerical statistics in the report can be challenging to interpret. Our objective is to streamline AGP analysis by utilizing Large Language Models (LLMs) and offering concise, narrative summaries that highlight significant patterns and information.

Method:

We utilized the BioMistral (Labrak et al, 2024) LLM. We prompted the LLM with numeric statistics and plots of data from 14 days of 10 pediatric T1D patients. The statistics were extracted using the AGATA toolkit (Cappon et al, 2023) to generate summaries. We tested various approaches and prompt formats to ensure outputs aligned with typical clinical analysis workflows. The same data were analyzed independently by two clinicians and used as input for the LLM. The LLM-generated summaries were compared to clinician analyses and evaluated by the same clinicians for accuracy and usefulness.

Result:

Comparative analysis revealed that the LLM-generated summaries effectively identified key information from the provided statistics. The LLM extracted the relevant information and delivered a brief text with enriched statistics, explanations, and, in some instances, suggestions on possible therapeutic actions. Clinicians rated the summaries as informative and indicated that such a tool could significantly reduce their analytical workload and streamline routine visits.

Conclusion:

BioMistral demonstrates a strong capability to identify and summarize critical information from AGP reports, mirroring clinician analyses. Further research is warranted to evaluate the tool's efficacy in clinical practice, with the potential to enhance efficiency and improve patient care.

Noninvasive Glucose Monitoring: Early Results from the GLUCOMFORT Tattoo Sensor

Cossu Luca, MSc; Cappon Giacomo, PhD; Francesco Prendin, PhD, Elisa Pellizzari, MsC; Alberto Maran, MD PhD; Federico Boscari, MD PhD; Stefano Lai, PhD; Erika Scavetta, PhD and Facchinetti Andrea, PhD

Department of Information Engineering, University of Padova Padova, Italy

cossuluca@dei.unipd.it

Objective:

Continuous glucose monitoring (CGM) sensors are crucial for effectively managing blood glucose levels in individuals with Type-1 diabetes (T1D). However, existing commercial CGM technologies present some limitations, such as high costs and the invasiveness of the measurement. To overcome these challenges, the GLUCOMFORT project aims to develop a noninvasive, affordable CGM system that uses a tattoo-based sensor with reusable components.

Method:

The GLUCOMFORT project (funded by the Italian Ministry of Research, ID: 2020X7XX2P) introduces a low-cost, multi-sensor array on a tattoo-like substrate. This array is connected to a reusable band that communicates with a mobile application for real-time data processing and visualization. The sensor utilizes reverse iontophoresis to extract glucose from sweat. The data collected by the reusable band are processed through calibration and filtering algorithms to convert raw readings into glucose concentrations, which are then displayed on a companion mobile application.

Results:

Preliminary in vitro testing of the tattoo-like sensor has demonstrated good sensitivity to glucose in the interstitial fluid. These initial tests have confirmed the feasibility and reliability of noninvasive glucose extraction and real-time data processing. Furthermore, the sensor has exhibited consistent performance, with limited variability observed across different test samples.

Conclusion:

The successful development of the GLUCOMFORT system could represent a significant advancement in the management of diabetes. By offering a low-cost, noninvasive CGM solution, the project holds great promise for improving patient compliance and health outcomes for individuals with T1D. The development of GLUCOMFORT sensor will continue with pivotal on-skin testing on healthy individuals and with a trial on a small group of people with T1D.

Nocturnal Hypoglycemia Detection

Pardiss Danaei, MAsc

Director of Machine Learning, Bio Conscious Technologies, Vancouver, BC, Canada

pardiss.danaei@bioconscious.tech

Objective:

Forecast the occurrence of nocturnal hypoglycemia in individuals with T1 Diabetes, within 4 hours. A nocturnal hypoglycemia event is when the blood glucose value is 70 mg/dL or lower when the individual is sleeping. For a nocturnal event we consider the window of 12 AM to 4 AM. Considering the importance of the 3 AM mark. For this project we assume a 12 AM bedtime since we do not have exact sleep time information for patients.

Method:

The method used is Deep Learning. We took advantage of Long Short Term Memory (LSTM) Neural Networks and trained a model to learn the glucose behavior 3 hours prior to bedtime and its correlation with the occurrence of a nocturnal hypoglycemia event from bedtime to 4 AM. The model performs binary classification to determine a positive or negative label for a nocturnal occurrence.

Result:

We calculated the Accuracy of the current model on 20 randomly selected patients. We achieved 81.6% with a high True Negative rate (93.5%). With further analysis, we believe that if we extend our dataset and balance our two classes we can increase the accuracy and robustness of the model.

Conclusion:

Identification of a nocturnal event is of utmost importance in patients with Diabetes since it can lead to coma or convulsions in severe cases. Our focus for this project is to initiate an approach to detecting the possibility of such an event so that the patient can take necessary measures. Our preliminary testing shows a model accuracy of 81.6%. Future development will include retraining the model with more accurate sleep time information.

A Pilot Study of Two Smartwatches with Alleged Glucose Monitoring Functionality

Manuel Eichenlaub, PhD; Stefan Pleus, PhD; Delia Waldenmaier, PhD; Guido Freckmann, MD

Institut für Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universität Ulm, Ulm, Germany

manuel.eichenlaub@idt-ulm.de

Objective:

Online retailers in Germany, but also the USA, have been offering smartwatches with an alleged functionality to non-invasively monitor glucose levels. In this context, the US FDA has issued a statement to not use such devices for diabetes therapy. The objective of this work was to assess the advertised glucose measurement capabilities of two smartwatches purchased through a large online retailer in Germany.

Method:

In a pilot study, a person with type 1 diabetes wore, on each wrist, the two acquired devices on two consecutive days. Simultaneously, measurements with a conventional continuous glucose monitoring (CGM) system were carried out. On the third day, the smartwatches were attached to a banana.

Result:

Each of the smartwatches recorded a distinct glucose profile that was almost identical from day to day, including the day the devices were attached to the banana. The glucose profiles of both smartwatches showed peaks of approximately 130 mg/dl and 160 mg/dl, respectively, following the usual timing of breakfast, lunch and dinner. Clinically relevant similarities to the simultaneously recorded CGM data on the first two days could not be identified.

Conclusion:

Based on the results of this pilot study it can be concluded that the two examined smartwatches possessed no actual functionality to measure glucose levels, which made further, more systematic examinations superfluous. Instead, it is suspected that the devices displayed a pre-programmed glucose profile that mimics the glucose levels of a person without diabetes. The results therefore support and illustrate the warning of the FDA and patients should not use such devices for therapy.

Performance of the 15-Day CareSens Air Continuous Glucose Monitoring System with Optional Calibrations

Manuel Eichenlaub, PhD; Stefan Pleus, PhD; Nina Jendrike, MD; Guido Freckmann, MD

Institut für Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universität Ulm, Ulm, Germany

manuel.eichenlaub@idt-ulm.de

Objective:

CareSens Air (i-SENS Inc., Seoul, Republic of Korea) is a novel continuous glucose monitoring (CGM) system featuring a 15-day sensor lifetime, 2-hour warm-up period and manual calibration, that recently received Conformité Européenne (CE) marking for adjunctive use. Since then, an updated calibration algorithm with a shortened warmup period of 30 minutes and optional, user-entered calibration was developed. The aim of this work was to assess the accuracy of the CGM system with the updated algorithm.

Method:

Using the updated algorithm, CGM data were generated from raw sensor data collected during a pivotal study with the CE-marked version of the system. This study included 50 adult participants with type 1 or type 2 diabetes and featured four, 8-hour in-clinic sessions with frequent measurement of venous comparator glucose concentrations and deliberate glucose level manipulations, spread over the 16-day study duration. Accuracy was characterized using the mean absolute relative difference (MARD) and agreement rates (AR) between CGM and comparator data, as well as a consensus error grid (CEG) analysis.

Result:

In total, 6066 CGM-comparator data pairs obtained from 50 sensors were analyzed. The overall MARD was 9.5%. The ARs were 83.1%, 91.8% and 99.6% for absolute/relative differences within ±15 mg/dl/%, ±20 mg/dl/% and ±40 mg/dl/% (comparator data ≤/> 100 mg/dl), respectively. The CEG showed 93.7%, 6.2% and 0.1% in zones A, B and C, respectively. The accuracy was stable over the sensor wear time (20/20 AR ranging from 88.8% to 93.0%).

Conclusion:

Compared to the CE-marked version of the CGM system, the accuracy could be substantially improved and the results support non-adjunctive use of the CGM system with the updated algorithm. Word count: 269 / 275 (max)

Full Data Transmission (Clinical and PROMS) from the Patient's Home to the Hospital to the National IT-Infrastructure and Back Again: Individually and Population-Based

Niels Ejskjaer, MD, PhD, Reporting on Behalf of the VBHC-PROMS Research Team

Aalborg University Hospital, Steno Diabetes Center, North Denmark, Denmark

n.ejskjaer@rn.dk

Objective:

Advancing delivery of person-centered and individualized diabetes care in order to improve outcomes and care for vulnerable individuals populations. Develop a seamless digital platform. Deliver a tool for improvement of quality of health care delivery and for decision support at strategic levels. Our VBHC-PROMS research and development team early on identified a significant need to advance delivery of person-centered and individualized diabetes care to improve outcomes, care for vulnerable populations, and overall reduce the burden of diabetes.

Method:

Through 2017-2021 we undertook a participatory multi-stakeholder research design process with systematic involvement of > 70 PWD and > 16 relatives using interviews, workshops, clinical & qualitative research, and through 2019-2021 our research and development team chaired a real-world 10-site PROMS study including 434 PWD and 34 HCPs showed high acceptability and demonstrating benefits related to the active engagement of PWDs focusing on person-centred individual diabetes care.

Result:

We have developed a scalable and efficient IT-solution to assist healthcare professionals (HCPs) and people with diabetes (PWD) work together to individualize medical diabetes management and integrate psychosocial and behavioral aspects of diabetes in a humanistic and person-centred way. Fully implemented at all four hospitals in the North Denmark Region, serving 600.000 citizens and 36.000 persons with diabetes. At this time 5500 PWD have answered a questionnaire in our four hospitals and 400 have answered their fourth questionnaire.

Conclusion:

It is feasible to implement a national digital PROMS questionnaire and a seamless intuitively and user-friendly national IT-platform, securing individualized treatment and delivering dynamic real-time individual data, assisting healthcare professionals and supporting data-driven management at all levels in the healthcare system.

First Data from Our Fully Digitalized Danish Diabetes PROMS Questionnaire: Identifying Unmet Psycho-Social Needs in Persons with Diabetes

Niels Ejskjaer, MD, PhD, Reporting on Behalf of the VBHC-PROMS Research Team

Aalborg University Hospital, Steno Diabetes Center, North Denmark, Denmark

n.ejskjaer@rn.dk

Objective:

Persons with diabetes (PwD) and our VBHC-PROMS-DIA research team developed the fully digitalized national Danish Patient-Reported Outcome (PRO) questionnaire (#DiaProfil). It is now fully implemented in all 4 hospitals in our region (uptake 600.000 inhabitants). DiaProfil identifies unmet psycho-social needs addressing appropriate individual action plans. This study is the first to psychometrically describe the magnitudes of need for selected psycho-social domains in a hospital population in Denmark.

Methods:

All PwD attending the outpatient clinic at Aalborg University Hospitals, completing the DiaProfil questionnaire (9 domains with 91 parameters) were included. The prevalence of each variable was determined using frequencies and percentages, while distributions were examined to clarify variable spread. T-tests were used to examine differences between variables.

Results:

1525 PwD (870 men and 655 women) with an average age of 50.9 years were included. Response rate 72 %. The average WHO-5 score was 63.7 points (national Danish non-diabetes score 68.5 points). We detected impacted sleep quality (40% disturbed sleep), sexual health (18% difficulty with intercourse), blood sugar regulation (39% report inadequate control), diabetes management (30% lacked confidence in their own diabetes management) and diabetes stress or depression (14,6 % in need of professional assistance).

Conclusion:

This study describes the prevalence of specific psycho-social need domains in adult PwD enabling individualized action planning. Specific domains in need were poor sleep quality, sexual health challenges, self-management confidence and undetected diabetes stress, all excerting a substantial impact on diabetes management, individual well-being and quality of life. Data is available for clinical use, hospital managers and the national IT-infrastructure servicing decisionmakers. Data may be downloaded by individual PwD's on intutively userfriendly dashboard. Full data transmission at all levels.

Assessing Beta-Cell Function in Type 2 Diabetes from Continuous Glucose Monitoring Data

Edoardo Faggionato, PhD; Chiara Dalla Man, PhD; Michele Schiavon, PhD

Department of Information Engineering, University of Padova, Padova, PD, Italy

edoardo.faggionato@unipd.it

Objective:

Beta-cell function (BCF) is usually estimated from plasma glucose, insulin, and C-peptide data collected after an IVGTT/OGTT in a hospitalized setting. The increasing use of continuous glucose monitoring (CGM) devices in individuals with type 2 diabetes (T2D) opens the door to monitoring BCF also in outpatient conditions.

Here, we propose and validate a Sensor Minimal Model (SMM) to quantify the BCF in individuals with T2D wearing CGM.

Method:

One hundred virtual subjects of the Padova T2D Simulator underwent a single-meal scenario (75g of carbohydrates). Plasma glucose, insulin, and C-peptide concentrations were frequently measured for 6 hours after the meal, and the (Oral Minimal Model) OMM was used to estimate the glucose rate of absorption (Ra^OMM) and the dynamic, static, and total disposition indices (DI_d^OMM, DI_s^OMM, and DI_tot^OMM). The CGM data were used to estimate DI_d^SMM, DI_s^SMM, DI_tot^SMM, and Ra^SMM using the proposed SMM.

Result:

Both the OMM and the SMM were able to well describe the data providing physiologically plausible parameters, estimated with precision. DI_s^SMM and DI_tot^SMM were highly correlated with DI_s^OMM and DI_tot^OMM (ρ=0.82 and ρ=0.84, respectively; p<10^-23), while a slightly lower, but still statistically significant correlation was obtained between DI_d^SMM vs. DI_d^OMM (ρ=0.66; p<10^-12). In addition, the area under the Ra at 2 hours obtained with SMM was highly correlated with the OMM-derived one (ρ=0.95; p=0).

Conclusion:

The proposed methodology can be used to assess BCF in individuals with T2D wearing a CGM sensor, and therefore, potentially, it can be employed in decision support systems for T2D management. Future work will include testing the method on real data of T2D with different stages of disease progression and therapeutic regimens.

Generating Digital Clones with the UVA/Padova T1D Simulator Identified on CGM and Pump Profiles via a Bayesian Approach

Edoardo Faggionato, PhD; Laya Ekhlaspour, MD; Bruce A. Buckingham, MD; Roberto Visentin, PhD

Department of Information Engineering, University of Padova, Padova, PD, Italy

edoardo.faggionato@unipd.it

Objective:

Digital cloning of real subjects can be a powerful tool for tailoring personalized therapies for individuals with diabetes. This can be realized by identifying a certain model on subject-specific data and using the estimated subject parameters in a simulation platform for the optimization of individual therapies through in silico trials. Here, we propose a digital-cloning method for subjects with type 1 diabetes (T1D) using real-life data collected from continuous glucose monitoring (CGM) sensors and insulin pumps (IP).

Method:

We used twelve single-day (7AM-11PM) individual datasets, consisting of CGM and IP profiles, as well as intakes of carbohydrates, available from 10 T1D subjects (age=14.0±16.7 years, BMI=20.7±4.6 kg/m²). Each individual dataset was used to obtain a digital clone by identifying the UVA/Padova T1D simulator model through a Bayesian estimator, assuming the a-priori parameter distribution implemented in the simulator for subject

generation. Parameters describing meal rate of appearance (Ra_meal) and insulin sensitivity (S_I) were allowed to vary during the day to capture different meal compositions and circadian variations.

Result:

The model well fit CGM traces, providing precise and physiologically plausible estimated parameters. Model internal states showed a physiological behavior as well. Ra_meal parameters were significantly different (pvalue<0.05) at breakfast compared to lunch and dinner, reflecting different meal compositions and/or overnight fasting. No significant differences were found in S_I parameters, likely due to the large inter-individual variability.

Conclusion:

The proposed digital cloning method can describe diurnal glucose profile of T1D subjects using minimally-invasive data, revealing underlying physiology aspects at individual level. Applied to larger (i.e., multiple-day/week) individual datasets, this method will open for tailored individual simulators to support more robust in-silico testing of personalized therapies.

Device Usage in Individuals with Diabetes for Self-Management

Sherecce A Fields, PhD, Rachel Smallman, PhD, Kianna Arthur, Samantha Philip, Vishaka Kalra, Kirsten Yehl, PhD, Erik Shoger, MBA, David Kerr MBChB, DM, FRCP, FRCPE

Texas A&M University, College Station, TX, USA

safields@tamu.edu

Objective:

This study aimed to understand device usage behaviors for individuals living with diabetes.

Method:

An online questionnaire was distributed to the dQ&A US Patient Panel (https://d-qa.com/). This comprised 3241 adults with diabetes: type 1 diabetes (N=1101), type 2 diabetes (T2D) on intensive insulin therapy (N=480), Type 2 diabetes on non-intensive insulin therapy (N=568), and Type 2 diabetes not on insulin (N=1092).

Result:

Approximately 14% of participants used some sort of smartphone application (app) to manage their diabetes. Participants reported using a wide variety of apps to track glucose, calories and steps. The most used devices included Apple Health, Dexcom Clarity, Fitbit, MyFitness Pal, and Samsung Health. Approximately 13% of participants used an app to log their carbohydrate intake and 33% used apps to track their exercise or activity. Approximately 19% of participants used apps to share data directly with their health care provider and 22% reported using apps to view and analyze past diabetes data. Participants reported that their healthcare providers advised them to use a variety of devices with most participants reporting being advised use DexCom devices. Approximately 50% of participants used continuous glucose monitors (CGMs), 35% used insulin pumps, and 6% used smart insulin pens or connected devices. About half of participants (47.4%) would like apps provided by CGM companies for viewing or interacting with their CGM data. Lastly, the majority of participants (72.4%) reported being advised to use finger stick blood glucose monitoring.

Conclusion:

Participants reported using a variety of technological & app-based solutions to manage their diabetes. A significant portion of individuals still use finger stick blood glucose monitoring to track their blood glucose.

Development of a Breath-Based Glucose Estimation Algorithm, A Progress Report

Frank Flacke PhD, Clarisse François-Marsal, MS; Theo Schiavi, MS; Laurent Cazals, None; Benoit Lepage, MD, PhD; Pierre Gourdy, MD, PhD, Sandrine Isz, PhD

BOYDSense, Toulouse, France

frank@flacke-consulting.com

Objective:

BOYDSense is developing a breath-based monitoring system for chronic conditions starting with glucose monitoring for people with type 2 diabetes (T2D). This is a post-hoc analysis of data from a proof-of-concept study evaluating the correlation of Volatile Organic Compounds (VOCs) in exhaled breath with glucose values. The aim of this analysis was to further improve the glucose estimation algorithm developed during the initial analysis of the data.

Method:

In this 2-phase study we enrolled 130 people with T2D, all subjected to a standardized meal challenge with predefined measurement timepoints at which 3 breath analysis and venous reference values were collected. In phase1 (n=100) data were collected for algorithm development. For the previous algorithm we split phase-1 data into 3 groups (training, verification, validation). In this step 100% of data from phase-1 were used to train the algorithm. The phase-2 data (n=30) stayed untouched and were again used for validation.

Result:

For analysis we used again the 612 data pairs, with 204 pairs for each breath. For the newly estimated average breath-based glucoses values the Consensus Error Grid Analysis (C-EGA) for T2D shows 86.3% in zone A and 99.5% in zone A+B with a stronger correlation for higher glucose values, and an improved MARD of 19.5% and a higher precision of 48.5% can be reported.

Conclusion:

By expanding the training data set the performance of BOYDSense breath analyzer demonstrated an improved correlation for high glucose values in the C-EGA, an improvement of MARD by 1.4% and an increase of precision by 4.4%. Two new studies, a clamp and a data collection study, providing more than 10,000 new data-pairs for further algorithm improvements, are underway.

Current Status of the Working Group on Continuous Glucose Monitoring of the International Federation of Clinical Chemistry and Laboratory Medicine

Guido Freckmann, MD; Robbert Slingerland, PhD; on behalf of the IFCC WG-CGM

Institut für Diabetes-Technologie Forschungs- und Entwicklungsgesellschaft mbH an der Universität Ulm Ulm, Germany

guido.freckmann@idt-ulm.de

Objective:

While current CGM systems have been shown to be safe and effective in diabetes therapy, individual CGM systems may report the same metabolic situation differently. To minimize this discordance, CGM performance studies must be standardized comprehensively. The Working Group on CGM (WG-CGM) of the International Federation of Clinical Chemistry and Laboratory Chemistry (IFCC) is working towards comprehensive standardization.

Method:

The performance of CGM systems is evaluated with respect to comparator blood glucose level measurements. While these comparator measurements should already cover the measuring range of the CGM system, concise requirements for the distribution of comparator data including rates of change are missing. Different comparator measurement methods have shown biases of up to 8% towards each other, and possibly more than 5% within the same brand, and concentrations in capillary and venous samples are physiologically different (often reported as 5 to 10% difference).

Result:

The WG-CGM recommends requirements for the distribution of comparator data to ensure adequate coverage of clinically relevant situations during testing. Furthermore, the WG-CGM recommends use of capillary samples, mainly because CGM results then are aligned better with self-monitored blood glucose results, which have to be performed in some situations even with factory-calibrated CGM systems. The accuracy of comparator methods should be established by showing compliance with analytical performance specifications. Comparator bias is recommended to be reduced through recalibration with respect to methods or materials of higher metrological order.

Conclusion:

Standardization of comparator data characteristics and comparator measurement procedures does not directly affect differences in calibration algorithms. However, once CGM readings are aligned with comparator results obtained from studies implementing standardized procedures, discrepancies between CGM readings from different systems will be minimized.

Feasibility of a Procedure to Produce Comparator Data for the Standardized Performance Evaluation of CGM Systems

Guido Freckmann, MD; Manuel Eichenlaub, PhD; Delia Waldenmaier, PhD; Stephanie Wehrstedt, PhD; Manuela Link, MD; Stefan Pleus, PhD; Sükrü Öter, MD; Nina Jendrike, MD; Maren Schinz, PhD, Cornelia Haug, MD

Institut für Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universität Ulm, Ulm, Germany

guido.freckmann@idt-ulm.de

Objective:

This study evaluated the feasibility of a recently proposed procedure for the manipulation of blood glucose (BG) levels for the standardized assessment of continuous glucose monitoring (CGM) system performance, which aims to produce a certain distribution of comparator BG levels representative of clinically relevant real-life situations.

Method:

24 adult participants with type 1 diabetes participated in three, seven-hour sessions with 15-minute capillary comparator BG level measurements. Participants consumed a carbohydrate-rich meal with a delayed insulin bolus to induce first hyper- then hypoglycemia associated with fast BG level changes, followed by stable BG levels in the normoglycemic range. Individual excursions were managed by an experienced physician using fast-absorbed carbohydrates, insulin boluses and mild exercise. BG levels and associated rates of change (RoC) were classified in the proposed dynamic glucose regions (DGR) “BG low” and “BG high” containing BG levels <70 mg/dl and >300 mg/dl, respectively, “Alert low” defined by BG≥70 mg/dl, RoC<-1 mg/dl/min and BG<70 mg/dl within 30 min at current RoC, and “Alert high” defined by BG≤300 mg/dl, RoC>+1.5 mg/dl/min and BG>250 mg/dl within 30 min at current RoC. Each DGR should contain at least 7.5% of BG-RoC pairs.

Result:

No adverse events related to the described procedure occurred. The percentages of BG-RoC pairs (n=1982) in all DGRs were ≥7.5%, with 10.9%, 13.5%, 9.8% and 7.5% in the “BG low”, “BG high”, “Alert low” and “Alert high” DGRs, respectively. The mean absolute BG level RoC was 1.44 mg/dl/min.

Conclusion:

The recommended comparator distribution could be achieved, thus demonstrating the feasibility of the procedure. However, to facilitate compliance with the recommendations, adaptations to the definitions of the DGRs and minimum percentages should be considered.

Needle-Free CGM Based on MagnetoHydrodynamics

Alejandro García Pérez, PhD; Laura Zschaechner, PhD; Reeta Saukkonen, MS; Kim Pettersson-Fernholm, MD-PhD; Edward Haeggström, Prof.

GlucoModicum, Helsinki, Finland

ag@glucomodicum.com

Objective:

To establish the correlation between glucose concentrations in interstitial fluid measured with Talisman needle-free continuous glucose monitor (CGM) and capillary blood glucose (CBG) concentrations measured with a reference meter in healthy volunteers and volunteers with type-2 diabetes. Talisman is a wearable device that enables needlefree continuous monitoring of glucose based on magneto-hydrodynamics (MHD).

Method:

Over 100 volunteers participated in the study (FIMEA-K485) authorized and monitored by the Finnish Medicines Agency (FIMEA) under the EU regulation. Glucose concentrations were concurrently measured with Talisman needle-free CGM and a reference CBG meter throughout a glucose tolerance test.

Result:

Talisman CGM demonstrated correlation (r² > 0.76) to reference CBG measurements which ranged from 77 to 467 mg/dl (4-26 mM).

Conclusion:

The achieved strong correlation evidences a high potential value of Talisman as a convenient and needle-free approach to glucose monitoring. Such a solution can help to reduce the burden of diabetes by increasing awareness, promoting earlier diagnosis, and facilitating the adequate treatment of diabetes globally.

Including Advanced Glucose Predictions into Diabetes Self-Management: The Accu-Chek® SmartGuide Predict App

Timor Glatzer, PhD; Magí Andorrà, PhD; Nils Babion, Dipl Ing; Hendericus Bos, Dipl Ing; Matthias Koehler, Dipl Ing; Yannick Klopfenstein, MSc; Eemeli Leppäaho, PhD; Patrick Lustenberger, MSc; Ajandek Peak, MSc; Christian Ringemann, PhD; Pau Herrero, PhD

Roche Diabetes Care GmbH, Mannheim, Germany

timor.glatzer@roche.com

Objective:

To support CGM users who continue to face challenges in achieving glycemic targets, experience diabetes distress and hypoglycemia fear, we propose the Accu-Chek^® SmartGuide Predict app, a CGM companion app that provides advanced glucose predictions aiming to inform users earlier about acute glycemic situations.

Method:

The Predict app, which has been developed following a user-centered design approach, is equipped with a 2-hour glucose forecast, a low glucose detection within 30 minutes, and a prediction of nighttime low glucose for bedtime interventions. Each prediction functionality is powered by an individual machine learning algorithm. The algorithms were trained on a proprietary dataset including people with T1D on multiple daily injections (MDI) insulin therapy (n=201). The evaluation of the algorithms encompassed three datasets, which included people with T1D on MDI (n=21) and pump therapy (n=201), as well as real-world data from people with T2D (n=59). In total, the evaluation covered ~64K subject days.

Result:

Across the three test datasets, the 2-hour forecast algorithm, at a prediction horizon of 30, 60, and 120 minutes, achieved a Consensus Error Grid A&B of 99.8%, 98.8%, and 96.6%, respectively. The low glucose detection algorithm achieved a sensitivity, specificity, and mean lead time of 94.5%, 97.5% and 19.3min, respectively. The nighttime low glucose detection algorithm achieved an accuracy, sensitivity, and specificity of 89.1%, 53.1%, and 94.8%, respectively.

Usability studies showed understanding of the app's functionalities and willingness to use its features. Task completion rate was 98.1%, indicating high success.

Conclusion:

Usability studies suggest high user satisfaction, engagement, and retention with the app. The consistent performance of the predictive algorithms on various cohorts from clinical studies as well as from real-world, offers reassurance for real-life efficacy.

Exploratory Study of Continuous Glucose Monitoring in the Epidural Space in Swine

Paul V. Goode, PhD; Mark A. Tapsak, PhD; Michael Talcott, D.V.M.; JP Thrower, PhD; Timothy L. Routh, M. Eng.; Stephen Tapsak; Samantha Wakil, PhD; Jose Garcia

GlucoTrack, Rutherford, New Jersey, USA

pvgoode@glucotrack.com

Objective:

Glucose can be potentially measured in various compartments of the body. This first-of-its-kind exploratory study evaluated epidural sensing as a viable method for continuous glucose monitoring.

Methods:

A preclinical study was conducted using a chronic porcine model to evaluate a novel long-term implant designed for the epidural space. The test device was comprised of an off the shelf spinal cord stimulation (SCS) lead that had been minimally modified by placing glucose sensing membranes atop 4 of the 8 existing stimulation electrodes. The resulting sensing lead and subcutaneous housing units were implanted in 4 swine, up to 56 days. Glucose levels were continuously monitored using the test device. Periodic glucose excursions were induced through oral glucose tolerance tests and intravenous glucose tolerance tests. Glucose measurements from the epidural space were compared to those obtained from blood and interstitial fluid using commercially available devices.

Results:

The results from glucose tolerance tests demonstrated that the test device reliably tracked glucose measurements similar to the blood and interstitial fluid. Results also indicate the swine tolerated the implant procedure well, lead placement was uncomplicated, and tissue healed with no adverse effects on the animals.

Conclusion:

The preliminary findings suggest that glucose sensing in the epidural space is feasible and reliable. This novel approach holds significant promise for simplifying diabetes management for patients when continuous glucose monitoring is integrated with an SCS device for people with Painful Diabetic Neuropathy. Further research is needed to assess long-term sensor performance.

Evaluating Machine Learning Methods for HbA1c Estimation from CGM Data: Impact of Data Pre-processing on Model Accuracy

Nada Haboudal, MS; Apoorva Karsolia, PhD; Cecilia Lee, MD, MS; Bhavesh Patel, PhD.

University of California San Diego, San Diego, CA, USA

nada.haboudal@gmail.com

Objective:

Hemoglobin A1c (HbA1c) measurement is crucial for diabetes diagnosis and management. Traditionally, HbA1c is measured via blood tests, but using regression models with data from Continuous Glucose Monitors (CGMs) can enable real-time, remote HbA1c monitoring. This study aims to improve HbA1c estimation accuracy by investigating two CGM data pre-processing methods for addressing non-numerical interval-censored data from Dexcom G6: first method interpolation to fill in interval-censored data,Second method removal of participant data with interval-censored data.

Method:

We tested the performance of ML methods for estimating HbA1c from CGM data using data collected using Dexcom G6 from 197 participants in the AI-READI study, a Data Generating Project funded by Bridge2AI, an NIH Common Fund Program. We evaluated four regression models: Linear Regression, Random Forest, XGBoost, and LSTM, using Mean Squared Error as the performance metric for both interpolated and non-interpolated datasets. For the interpolated dataset, interval-censored CGM values (non-numerical Min and Max values) were replaced with the local minimum and maximum values for each study subject. For the non-interpolated dataset, participants (n=63) with intervalcensored CGM data were removed. Cross-validation determined the statistical significance of model performance differences.

Results:

Random Forest demonstrated superior performance in predicting HbA1c levels with the lowest MSE for both interpolated (0.45) and non-interpolated (0.38) data. Random Forest and XGBoost showed no statistically significant difference between interpolated and non-interpolated datasets, while the other models showed improved accuracy on non-interpolated data.

Conclusion:

The study concludes that the choice of pre-processing method significantly impacts the accuracy of HbA1c predictions.The results showed that using non-interpolated data, which involves removing samples with nonnumerical interval-censored CGM data, can yield more accurate predictions than interpolating these values.

An Investigation of Cutaneous Reactions to Continuous Glucose Monitors Using a Patient-Centered Survey Tool

Emilee Herringshaw, MD, MBA, Felix Raimundo, PhD, Cheryl Berry, RN, MS, CDCES, Samir Malkani, MD, David Harlan, MD, Wei-Che Ko, MD

UMass Chan Medical School, Department of Medicine[HE1], Worcester, MA, 01655, USA

emilee.herringshaw@umassmed.edu

Objective:

Cutaneous reactions are occurring in patients utilizing continuous glucose monitors (CGMs). A patient centered survey was created to provide insight on trends and management related to CGM reactions.

Method:

A survey was made available at clinics through the UMass Diabetes Center of Excellence and online through social media. Participant demographics, medical history and device use characteristics (CGM model, occurrence of a reaction, frequency of reaction, frequency of early device removal, reaction severity (1:5), reaction evolution over time, interventions and impact of interventions) were recorded.

Result:

Participants (n=316) were 35.6 +/- 19.5 years old on average. The majority were white (86.0%, 270/314), nonHispanic (89.2%, 272/305) and female (68.1%, 213/313) with Type 1 Diabetes (85.8%, 271/316). CGMs used included: Libre (12.0%, 38/316), Dexcom g6 (68.7%, 217/316), Dexcom g7 (12.3%, 39/316), Eversense (5.4%, 17/316) or other (1.6%, 5/316). Reactions occurred in 59.3% (185/312) of patients at a frequency of rarely (25.1%), sometimes (27.4%), often (43.4%) and always (25.1%). Early removal of the device occurred at a frequency of never (4.6%), rarely (30.6%), sometimes (30.6%), often (11.6%) or always (22.5%). Reaction severity was normally distributed across different body sites. Patients reported reactions got worse 38.5% (70/182) of the time. Interventions were attempted by 77.7% (146/188) of participants, resulting in no change for 39.7% (56/141), the reaction becoming worse for 34.8% (49/141), much worse for 0.7% (1/141) or better for 24.8% (35/141).

Conclusion:

Patients with diabetes are developing reactions to CGMs. Enhanced awareness of factors associated with reactions as well as understanding about interventions and implications can inform guidance regarding the development and management of reactions.

Cutaneous Reactions to Continuous Glucose Monitors: Predictive Factors of Reaction Severity

Emilee Herringshaw, MD, MBA; Felix Raimundo, PhD; Cheryl Berry, RN, MS, CDCES; Samir Malkani, MD; David Harlan, MD; Wei-Che Ko, MD

UMass Chan Medical School, Department of Medicine, Worcester, MA, 01655, USA

emilee.herringshaw@umassmed.edu

Objective:

Patients are experiencing cutaneous reactions to continuous glucose monitors (CGMs). Our study sought to evaluate factors associated with reaction severity to inform which patients were at risk of developing severe reactions upon starting a CGM.

Method:

A survey was offered to people wearing CGMs, at clinics within the UMass Diabetes Center of Excellence and through social media outreach. Data was collected on participant demographics and medical history, device model, device use characteristics and reaction severity on a scale of 1 to 5. Factors were analyzed for their association with severity (no reaction (1), mild-moderate reaction (2-3), severe reaction (4-5)).

Result:

The factors associated with increased reaction severity included a pre-existing diagnosis of contact dermatitis (p<0.001), type of diabetes (Type 1 Diabetes vs. Type 2 Diabetes, p<0.01) and age (greater severity in younger patients, p<0.001). Severity may also be associated with affordability, given limitations on device selection (trend to significance p = 0.08) Severity did not vary with total duration of device use or days per CGM session. Analysis was limited by the information collected and self-reported nature of the survey. Furthermore, most of the analyses were performed in the context of Dexcom g6 users (71%, 109/53) given the device use of survey respondents.

Conclusion:

Although it is known that people with diabetes are experiencing reactions to CGMs, there is limited understanding of who will develop a reaction that interferes with device use or significantly impacts quality of life. Here, we investigate which patients may be at risk for developing a severe reaction which may help inform device selection, tailor anticipatory guidance to reduce risk of a severe reaction or improve counseling regarding reaction management.

Continuous Glucose Monitoring with an Osmotic-Pressure Based Continuous Glucose Sensor – Human Pilot Study Results and Next Development Steps

Joacim Holter, Jo Amundstad, Sievert Nerhagen, Nicole Thomé, Boris Stamm, Andreas Pfützner,

Lifecare AS, Bergen, Norway

joacim.holter@lifecare.no

Background:

The osmotic-pressure-based Sencell continuous glucose sensor technology (Lifecare AS, Bergen, Norway) is expected to be implanted into the s.c. tissue for long-term usage and to employ wireless energy and data transfer.

Method:

In a first clinical proof of concept study in humans, a wired version of the core sensing technology was embedded into a 4 mm needle and inserted into the abdominal subcutaneous tissue for up to three days of use. The raw data was retrospectively analyzed after one-point calibration and minor trend correction in comparison to the Statstrip blood glucose meter and commercially available CGM glucose sensors.

Results:

Nine subjects (6 female, 3 male, age: 49±11 years, including 1 subject with type 1 diabetes) delivered a total of 261 direct comparator data-points (vs. Statstrip blood glucose meter) during repeated meal experiments. The osmoticpressure sensor signal reached an overall MARD of 9.6% in comparison to StatStrip. In the retrospective analysis with the newly developed algorithm, 90.8 % and 9.2 % of the datapoints were lying in zones A and B of the consensus error grid, respectively.

Conclusions:

After development of a first algorithm to translate sensor signals into a glucose concentration, the osmotic-pressure based continuous glucose sensor was shown to track s.c. glucose concentrations in a comparable manner as commercially available needle sensors. In consequence, a next development step was initiated to integrate suitable electronics and energy sources for wireless data and energy transfer. First prototypes have been manufactured and are currently investigated for their longevity of use in a veterinary study with dogs with a minimum duration of use of 3 months.

Enhancing Safety in Type 1 Diabetes: A Decision Tree Approach to Insulin Pump Malfunction Detection

Elena Idi, MSc; Andrea Facchinetti, PhD; Giovanni Sparacino, PhD; Del Favero, PhD

University of Padova, Padova, Italy

elena.idi@studenti.unipd.it

Objective:

Insulin pumps have shown their effectiveness in improving the quality of the treatment for individuals with type 1 diabetes. Nevertheless, safety concerns may arise due to infusion set failures, that can potentially interrupt insulin delivery, exposing the patients to the potential risks associated to hyperglycemia and even ketoacidosis. This work proposes a novel approach for detecting insulin pump occlusions using a decision tree classifier.

Method:

Data of 100 virtual subjects were simulated over 30 days using the UVA/Padova T1D Simulator, with each subject experiencing one pump occlusion at midnight of a random day. From this data, 15 features were extracted to highlight the abnormal conditions related to the pump occlusion. Feature ranking was performed using the Gini importance metric and resulted in the selection of 4 key features from the original 15 to be used as inputs for the decision tree classifier.

Result:

On the test set, that consists of 50 previously unseen subjects, decision tree detects the 78% of the insulin pump occlusion, while raising on average 0.09 false positives per day, equivalent to less than 3 false alarms in one month. The average detection delay is about 230 minutes.

Conclusion:

Despite the simplicity of decision trees, their interpretability combined with the promising results achieved in the simulated scenario, suggest their potential for their application to detect insulin pump occlusion in real-world data.

Feasibility of Multi-Metabolic Continuously Monitoring

Brian Kannard, MBA; Bradley Liang, MS; Katherine Wolfe, MS; Ben Pagliuso, BS; Bella Nava; BS; Blake Morrow, MS, MBA; Elizabeth Esponoza, BS; Kaitlynn Olczak, PhD; Rajiv Shah, MS

PercuSense, Inc., Valencia, California, USA

brian.kannard@percusense.com

Objective:

Continuous glucose monitors (CGM) are used to enable automated insulin delivery while improving clinical outcomes in people with diabetes. Multi-analyte continuous monitoring combines the advantages of CGM with monitoring other key metabolites important to managing diabetes. Resting lactate elevation is reported to be a marker of insulin resistance and diabetic kidney disease and has been strongly associated with higher incidences of cardiac disease, cancer, and immune system dysfunction. Additionally, lactate can measure exercise and activity as a marker and be useful in optimizing exercise protocols (e.g., Zone 2 exercise). Establishing the feasibility of a multimetabolic sensor could enable future products that can improve the clinical outcomes of people with diabetes.

Method:

PercuSense has developed a multi-analyte continuous sensing platform capable of measuring up to three analytes with a sensor and wearable size similar to commercially available CGM systems. A previous, larger system was tested in humans. This study presents results from an improved sensor, new insertion tool, and miniaturized wearable tested in diabetic rats. Glucose and lactate levels were modulated via meals and exercise, and meter blood glucose and lactate samples were taken periodically to evaluate sensor accuracy.

Result:

The sensors accurately measured glucose and lactate levels at rest and during meals and exercise over multiple days without crosstalk between signals. The glucose sensor MARD is 9.91% and the lactate 30% agreement rate is 82.5%. This study establishes the feasibility of combined glucose and lactate sensing with a commercially viable miniaturized system.

Conclusion:

Multi-analyte sensing is an important tool in providing a more complete picture of metabolic health and aiding people in the long-term management of diabetes. PercuSense has developed a system capable of multi-metabolic sensing.

Reduced Healthcare Provider Burnout Associated with Use of Spotlight-AQ: Focus on Physical, Mental and Social Wellbeing

Ryan Kelly, BSc, MBA; Hermione Price, MD, PhD; Michael Cummings, MD; Amar Ali, MD; Mayank Patel, MD; Ethan Barnard, BSc; Katharine Barnard-Kelly, PhD

Spotlight-AQ, Fareham, Hampshire, United Kingdom

ryan@spotlightaq.co.uk

Objective:

Burnout, including emotional exhaustion, depersonalisation and lack of personal accomplishment, affects >50% of physicians and nurses and is estimated to cost £6000 per employed physician per year. It is associated with high rates of sick leave, medical errors, poor job satisfaction, greater staff turnover and poorer patient satisfaction and outcomes. Furthermore, burnout increases the risk of suicide, alcohol, tobacco and substance abuse. Spotlight-AQ is a digital health platform designed to improve routine diabetes visits for patients and healthcare professionals (HCPs) through use of a biopsychosocial pre-clinic assessment and mapped resources. We aimed to assess whether Spotlight-AQ was associated with reduced burnout for HCPs.

Methods:

A multi-centre, parallel group, randomised clinical trial evaluated the efficacy and cost-effectiveness of Spotlight-AQ as well as impact on HCP burnout. The Maslach Burnout Inventory was used to assess emotional exhaustion, depersonalisation and personal accomplishment for participating HCPs at baseline and again at 6 months follow-up.

Results:

18 HCPs participated. Burnout was reduced overall and across all subscales ie reduced depersonalisation (-1.61 (0 to -12)), improved sense of personal achievement (4.61 (-4 to 13)) and reduced emotional exhaustion (-2.28 (0 to -18)). Taken in context of patient participant results showing reduced HbA1c (mean 6mmol/mol), shorter visit length (14%, 0.5-6 mins across 15 & 30 min visits) and cost-effectiveness (68% likelihood of cost-effectiveness, £98 per patient cost saving compared to controls), the intervention improves both the patient and HCP experience during the consultation. Doctors (n=10) reported greater improvements than nurses (n=8) across reduced depersonalisation -2.88 (1 to -12) vs -0.67 (1 to -3) and emotional exhaustion -3.24 (5 to -18) vs -1.67 (1 to -8) but not on improved sense of pesonal achievement 4.63 (-4 to 13) vs 5.11 (0 to 13).

Conclusion:

Spotlight-AQ was associated withreductions in burnout and depersonalisation for HCPs, as well as greater sense of personal achievement. Burnout is affected by many factors within and beyond the workplace and this study could not demonstrate causal relationship. Further research is required.

Spotlight-AQ Versus Usual Care for Adults with Diabetes: A Trial-Based Complete Case Cost-Effectiveness Analysis

Ryan Kelly, BSc, MBA; Hermione Price, MD, PhD; Michael Cummings, MD; Amar Ali, MD; Mayank Patel, MD; Ethan Barnard, BSc; Katharine Barnard-Kelly, PhD

Spotlight-AQ, Fareham, Hampshire, United Kingdom

ryan@spotlightaq.co.uk

Objective:

Spotlight-AQ is a digital health tool for healthcare professionals (HCPs) that maps patient concerns (psychological wellbeing, therapy, self-management, and social support) and provides evidence-based resources to address them. It is associated with reduced overall HCP burnout and across all sub-domains (burnout, depersonalisation and emotional exhaustion). We aimed to evaluate cost-effectiveness of Spotlight-AQ for adults with diabetes in England.

Methods: A complete case within-trial cost-effectiveness analysis was performed (perspective: health care system; price year: 2021/22; time horizon: 6-months). This study was embedded within the Spotlight-AQ pivotal multicentre, parallel group, randomized controlled trial. Adults with diabetes were recruited within primary and secondary care. Intervention group participants completed the Spotlight-AQ preclinic assessment before each routine outpatient visit. Control group participants followed care as usual. Health status was measured by the EQ-5D-5L (value set: EQ-5D3L UK tariff) to calculate quality-adjusted life years (QALYs). Downstream resource use was measured by the Client Service Receipt Inventory. Total costs and QALYs per arm were calculated by baseline-controlled seemingly unrelated regression with bootstrapping. HbA1c and visit length data were collected.

Results: There were 98 participants (49 intervention). Total costs: £243 for intervention arm and £230 for control arm (incremental cost: £13). Total QALYs: 0.362 for intervention arm and 0.358 for control arm (incremental QALY: 0.004). The Spotlight-AQ intervention dominated usual care; 68% probability of being cost-effective (threshold value: £30,000 per QALY gained). Intervention group achieved HbA1c improvement (0.6%, 6 mmol/mol) vs control (0.3%; 3 mmol/mol), p≤0.001; and shorter visit lengths (0.5-6 mins, mean 4.1 mins) vs (-0.9 to +1.3 mins) control. P≤0.001.

Conclusion: The complete case analysis provides evidence that Spotlight-AQ is cost-effective in routine care of people with diabetes.

Real-World Survival of the Medtronic Extended Infusion Set in the United States

Justin Kiehl BA; Tim Kwa, PhD; Gina Zhang, PhD; Venkat Putcha, PhD; Amanda Kinnischtzke, PhD; Sarnath Chattaraj, PhD; and Ohad Cohen, MD

Medtronic Diabetes, Northridge, CA, USA

justin.kiehl@medtronic.com

Objective:

The Medtronic extended infusion set (EIS) is the first and only infusion set approved for up to 7-day use and has proven clinical safety and efficacy. The purpose of this retrospective analysis was to estimate EIS survival based on United States (US) real-world data (RWD) uploaded by MiniMed™ 780G system users at an aggregate level.

Method:

Data from the pivotal trial (n=253 users), which contained end-user-reported set changes along with CareLink™ data from each user’s device, was used to train and validate both a rule-based and supervised machine learning model to determine the most effective model for estimating infusion set wear duration. The models were subsequently applied to RWD (n=616 identified as EIS users by having placed at least 6 orders of infusion sets, 80% or more of which were EIS, between May 2023 and May 2024) to estimate EIS survival in the field.

Result:

Among the tested models, the XGBoost classification model was most accurate (91% accuracy) for identifying infusion set changes in the clinical dataset, using the tubing fill amount, cannula fill amount, post-rewind reservoir volume, difference between time of rewind and estimated empty reservoir, and a low insulin flag (user-defined, typically 10 U to 20 U). Applying the model to RWD, the resulting median wear duration was 7.02 days with a mean of 7.81 ±3.87 days.

Conclusion:

With median wear duration of 7.02 days, the US RWD analysis revealed that EIS performance is similar to that in the pivotal trial (7.38 days). Combined with recent analyses showing a lower occlusion rate without compromising glycemic outcomes, the EIS solves one of the unmet needs in pump management thereby substantially reducing user burden.

Time to Moderate and Severe Hyperglycemia and Ketonemia Following an Insulin Pump Occlusion

David C. Klonoff, MD, FACP, FRCP (Edin), Fellow AIMBE; Alessandra T. Ayers, BA; Cindy N. Ho; Chiara Fabris, PhD; María Fernanda Villa-Tamayo, MS; Eleanor Allen, MD; Eda Cengiz, MD, MHS, FAAP; Laya Ekhlaspour, MD; Jenise C. Wong, MD, PhD; Lutz Heineman, PhD; Michael A. Kohn, MD, MPP

Diabetes Research Institute, Mills-Peninsula Medical Center, San Mateo, California, USA

dklonoff@diabetestechnology.org

Introduction:

Insulin pump therapy can be significantly impacted by blockages that obstruct insulin flow, leading to impaired insulin delivery. This can result in elevated blood glucose (BG) levels and an increase in beta-hydroxybutyrate (BHB) ketone levels. The duration before serious effects of an occlusion manifest depends on the rate of insulin flow and individual variability.

Method:

We conducted a PubMed search for relevant studies published in English from January 1982 to July 2024. Our aim was to estimate the rate of increase in glucose and/or BHB levels following a disruption in subcutaneous insulin infusion in individuals with type 1 diabetes (T1D). Additionally, we simulated glucose level increases in a virtual cohort of 100 adults with T1D after cessation of continuous subcutaneous insulin infusion (CSII).

Results:

We identified seven pertinent studies where BG and BHB (in 6 out of 7 studies) were measured after CSII suspension, serving as a model for occlusion. After the first 60 minutes of stable BG and BHB concentrations postsuspension, the average rates of increase were 0.61 mg/dL/min for glucose and 0.0034 mmol/L/min for BHB. The estimated mean times to reach moderately/severely elevated BG levels (300/400 mg/dL) or BHB (1.6/3.0 mmol/L) were 5.9/8.7 hours for BG and 8.8/15.7 hours for BHB, respectively. The simulation model projected that virtual subjects would experience moderately/severely elevated BG levels (300/400 mg/dL) after 9.25/12, 6.75/8.75, and 4.75/5.75 hours, corresponding to small (5th percentile), medium (50th percentile), and large (95th percentile) hyperglycemic responses to the interruption.

Discussion:

The calculations from clinical studies and simulation models indicated similar timeframes for the onset of moderate to severe metabolic risks following CSII interruption. Typically, moderate or severe metabolic complications can occur within approximately 6 to 12 hours after a CSII occlusion.

Downtime Impact Factor evaluates Downtime in Automated Glucose Clamps

Mareike Kuhlenkötter, MSc; J. Hans DeVries, MD; Carsten Benesch, PhD

Profil, Neuss, Germany

Mareike.kuhlenkoetter@profil.com

Objective:

The glucose clamp technique is the gold standard to describe pharmacokinetic and pharmacodynamic parameters of blood glucose (BG) lowering agents. Established glucose clamp quality parameters, such as control deviation and precision, characterize the clamp quality solely based on quality of achieved BG control. In particular, current clamp quality parameters are independent of the glucose infusion rate of change induced by the glucose lowering agents. We developed the Downtime Impact Factor, considering rate of change of glucose infusion, to determine the impact of technical downtimes on the quality of glucose clamp data. In periods of technical downtime, staff takes over with manual clamping.

Method:

Simulated clamp data were used to determine the effect of downtimes within glucose clamps. Episodes of different length shut-offs of automated BG regulation were implemented at time points with different slopes in the glucose infusion rate profile of the BG lowering agent. Based on these data we determined how these episodes affect clamp results. Finally, the method was retrospectively applied to real-world clamp data.

Result:

The Downtime Impact Factor was applied to data of an automated glucose clamp study with 26 subjects in a crossover study design (625 hours of automated clamp data). Overall downtime was 32.75 hours (5.2%) with a median downtime duration of 5 (IQR 5) minutes. The impact of downtime depended on its duration and the actual rate of change in Glucose Infusion Rate.

Conclusion:

A new quality parameter named Downtime Impact Factor was validated to more precisely describe the impact of technical downtimes on automated glucose clamp outcomes.

Evaluating Perplexity and Glucose Level Impact on State-Of-The-Art Generative Pre-Trained Transformer (GPT) Model to Predict Glucose Values at Different Time Intervals

Abhimanyu Kumbara, MS, MBA; Junjie Luo, MS; Anand K. Iyer, PhD, MBA; Mansur E. Shomali, MD; and Guodong “Gordon” Gao, PhD

Welldoc Inc., Columbia, Maryland, USA

akumbara@welldocinc.com

Objective

AI has the potential to make CGM data more helpful for people living with diabetes. We have been developing GPT models to predict CGM trajectory at different time horizons. In this study, we evaluated the performance of our GPT model across two prediction contexts: (1) the prior 24 hours glucose out-of-range frequency and (2) model perplexity.

Method

A GPT model to generate CGM trajectories at 30-minute, 60-minute, and 2-hour time intervals was created using a real-world data set from 592 CGM users. Glucose out-of-range frequency was then classified into 5 categories: very low, low, medium, high, and very high. A perplexity score was also calculated. Perplexity is a measure of prediction uncertainty, and in this context denotes how “surprised” a model is by a given glucose value input, based on the data the model was trained on.

Result

As the prior 24-hour glucose out-of-range frequency increased, our models’ root mean square error (RMSE) also increased. The largest increase in RMSE was noted when going from medium out-of-range category to high out-ofrange category, with RMSE for 2-hour CGM generation increasing from 28 mg/dL to 33 mg/dL respectively. 2-hour CGM generation RMSE also increased as the model perplexity increased from very low to very high.

Conclusion

Our evaluations of the GPT model used to generate glucose trajectories showed that as the glucose out-of-range frequency and perplexity increased, model performance decreased. Just as large language models need to be finetuned to fit specific language generation tasks, large glucose models such as ours will also need to be further finetuned, based on different glucose profiles of the population.

Identifying Severe Insulin-Deficient Type 2 Diabetes Subphenotype in Electronic Health Records from USA

Zhongyu Li, MSPH; Star Liu, MS; Jithin Sam Varghese, PhD, MTech

Nutrition and Health Sciences Doctoral Program, Laney Graduate School, Emory University Atlanta, GA, USA

zhongyu.li@emory.edu

Objective:

To identify severe insulin-deficient diabetes (SIDD), a novel type 2 diabetes (T2DM) subphenotype associated with high risk of microvascular complications, using machine learning and routine clinical variables in electronic health records (EHR).

Methods:

Using unsupervised clustering (k-means) of newly diagnosed T2DM patients (n = 3,771) from six harmonized US cohorts (ARIC, CARDIA, DPP, DPPOS, JHS, MESA), we replicated four novel T2DM subphenotypes based on five variables (age at diagnosis, body mass index [BMI], HbA1c, homeostatic model assessment 2 indices). Next, we fit different supervised classification models on a 70% training dataset to differentiate between SIDD and other T2DM subphenotypes (severe insulin-resistant [SIRD], mild age-related [MARD], mild obesity-related [MOD]) using nine routinely collected clinical variables (age at diagnosis, BMI, systolic and diastolic blood pressure, HbA1c, LDL, HDL, triglycerides, triglyceride-to-HDL ratio). Performance (sensitivity, specificity, AUC) was assessed using a 30% hold-out test dataset. Finally, the model was applied to EHRs of newly diagnosed T2DM (n = 727,094) from the Epic Cosmos Research Platform, identified using the SUPREME-DM computable phenotype.

Results:

The pooled cohort sample was on average aged 61.7 (SD:12.8) years, 60.2% female, 42.4% Non-Hispanic Black,

45.5% Non-Hispanic White, and 8.7% Hispanic. Patients were categorized into four groups: SIDD (6.2%), MARD (46.8%), MOD (39.3%), and SIRD (7.7%). The multivariable logistic regression for detecting SIDD achieved high performance in the test set (sensitivity: 0.86, specificity: 0.99, AUC: 1.00). Application of the trained model identified 17.9% of T2DM patients in Epic Cosmos as SIDD, with wide variation between states (Nebraska: 7.8%; District of Columbia: 24.4%).

Conclusion:

Model-based classification of SIDD achieved using routine clinical variables offers significant potential for integrating precision medicine into clinical workflows.

Performance of a Novel Continuous Glucose Monitoring Device in People with Diabetes

Julia K. Mader, MD; Delia Waldenmaier, PhD; Julia Rötschke, PhD; Gerhard Vogt, PhD; Michael Angstmann, PhD; Katrin Müller, PhD; Sandra Moceri, PhD; Thomas Forst, MD; Guido Freckmann, MD

Division of Endocrinology and Diabetology, Medical University of Graz, Graz, Austria

julia.mader@medunigraz.at

Objective:

Continuous glucose monitoring (CGM) improves glycemic control in people with diabetes (PwD). The availability of multiple highly accurate systems allows PwD and healthcare professionals to choose the best option for the needs of each individual. This study evaluated the performance of the novel Accu-Chek® SmartGuide device.

Method:

An open-label, single-arm, prospective, non-randomized, multicenter study was conducted with 48 adults with type 1 (n=40) or type 2 (n=8) diabetes on insulin therapy. Participants wore three sensors at the back of the upper arm for up to 14 days. During in-clinic visits, glucose manipulations into hypo- and hyperglycaemia were performed and CGM data was compared to capillary glucose measurements (Accu-Chek® Guide). The primary endpoint was the percentage of CGM readings within ±20mg/dL or ±20% (for blood glucose (BG) values <100mg/dL and ≥100mg/dL respectively) of the comparator (20/20 AR). Further evaluations included the mean absolute relative difference (MARD) and the 20/20 AR in different glucose ranges and across the wear time.

Result:

Data from 139 sensors were analyzed. During in-clinic visits, the 20/20 AR was 90.5%, and the MARD 9.2%. For BG values <70mg/dL, 70-180mg/dL; and >180mg/dL, the 20/20 AR was 94.3%, 89.0% and 92.5%, respectively. At the beginning, middle, and end of wear time, the 20/20 AR was 92.8%, 91.5%, and 85.9%, respectively. The 14-day sensor survival probability was 82.4%. Pain and bleeding after sensor insertion were within the expected range.

Conclusion:

The Accu-Chek® SmartGuide device showed a good performance over the entire measurement range, especially in the low range, and the whole wearing time of the sensors with respect to capillary BG measurements. This study supports non-adjunctive use after the initial calibration of the device.

The Effects of Melatonin Supplementation and Continuous Glucose Monitoring on Glycemic Control, Glucose Variability, and Sleep Quality in a Patient with Type 2 Diabetes: A Case Study

Masab A. Mansoor, DBA; Andrew F. Ibrahim, BS; Nicholas Kidd, MD

Edward Via College of Osteopathic Medicine, Monroe, Louisiana, USA

mmansoor@vcom.edu

Objective:

Melatonin supplementation has been studied for its potential benefits in improving sleep quality and glycemic control in type 2 diabetes. Continuous glucose monitoring (CGM) provides real-time data on glucose levels and variability, which can help improve diabetes management compared to non-continuous monitoring. This case study investigates the effects of melatonin supplementation and the subsequent addition of CGM on glycemic control, glucose variability, sleep quality, and self-monitoring in a 55-year-old patient with a 20-year history of type 2 diabetes and coronary artery disease.

Method:

The 55-year-old male patient has been self-monitoring for 10 years. Melatonin supplementation (3mg/night) was initiated, and after one year, CGM was added. Data on glycemic control (HbA1c), glucose variability (coefficient of variation, CV), sleep quality (Pittsburgh Sleep Quality Index, PSQI), and self-monitoring frequency were collected at baseline, 1 year (post-melatonin), and 2 years (post-CGM).

Result:

At baseline, the patient's HbA1c was 8.2%, CV was 36%, PSQI score was 11 (poor sleep quality), and selfmonitoring frequency was 3 times/day. One year after starting melatonin, HbA1c decreased to 7.6%, CV reduced to 32%, PSQI score improved to 7 (borderline sleep quality), and self-monitoring frequency remained unchanged. Two years after baseline and one year after adding CGM, HbA1c further decreased to 6.8%, CV reduced to 28%, PSQI score was 5 (good sleep quality), and self-monitoring frequency increased to 5 times/day.

Conclusion:

Melatonin supplementation may have improved glycemic control, glucose variability, and sleep quality. Adding CGM may have further improved these parameters and increased self-monitoring frequency. This case study suggests that melatonin and CGM may synergistically affect diabetes management.

Real-Life Use of Text Messaging For Delivery of DSMES for People with Type 2 Diabetes

Melinda D. Maryniuk, MEd, RD, CDCES; Ansley Dalbo, BA

Lead Diabetes Care and Education Specialist, Diabetes – What to Know, Dennis, Massachusetts, USA

melinda@melindamaryniuk.com

Objective:

Data from the CDC shows less than 7% of adults with diabetes receive diabetes self-management education and support (DSMES) services within the first year of diagnosis. According to the 2022 National Standards for DSMES, evidence supports DSMES delivery via non-traditional methods including text messages. Real world applications of text messaging for DSMES are relatively rare, however.

We created and launched a program for ReliOn^® customers with type 2 diabetes that delivers DSMES through a weekly text message. The objective was to expand the number of people receiving diabetes information and support, and to evaluate the impact of such a program in the real world.

Method:

After launching a successful pilot phase of the program in 2023, planning began for a national roll-out in January 2024. People who purchased a ReliOn^® meter were able to opt-in to the program via a brochure in the meter box at the end of April 2024 and the program was launched for existing ReliOn^® customers via ReliOnBGM.com in July 2024.

Result:

Over 3,700 individuals have enrolled since April 2023. The cumulative unsubscribe rate for the program is ~20%, resulting in ~2,950 active participants currently enrolled.

Delivery rates are at 98% and click-through rates for the program range between 20% - 40% (average of ~30%) depending on the message. (This compares to email benchmarks of 37% open rates and 1.8% click through rates.) Pilot study evaluations revealed 98% responding that the program was helpful in managing their diabetes, and 77% reported feeling motivated to check glucose more often.

Conclusion:

Text messaging is an effective way to engage with patients with far higher open and click-through rates than email.

Novel Wearable Ring with Medical Grade Pulse Oximetry and Other Wellness Metrics

John J Mastrototaro, Ph.D.; Michael Leabman; Kim Tompkins

Movano Health, Inc., Pleasanton, CA, USA

jmastrototaro@movano.com

Objective:

The objective of this study was to evaluate the reflectance PPG pulse oximetry feature of a novel medical device in the form of a wearable ring, EvieMED, which also provides comprehensive health and wellness information, including monitoring of heart rate, HRV, respiration rate, skin temperature, activity, sleep, and accommodates logging capabilities for menstrual cycle, mood, energy, and other symptoms.

Method:

A standardized hypoxia study was carried out at UCSF where 11 healthy volunteers with a broad range of skin tones breathed gas mixtures to lower oxygen levels from their normal 95-100% down to 70% oxygen in a stepwise fashion. At each plateau, three arterial blood gas samples (SaO₂) were obtained to provide the reference values. Subjects were outfitted with four EvieMED ring test devices of the appropriate ring size, two worn on fingers and two held on fingertips, as well as two hospital grade pulse oximeters. UCSF collected the data from all devices and conducted an accuracy analysis.

Result:

The result showed that the four EvieMED test locations achieved a similar level of accuracy of 2.39% to 2.53% RMSE, well within the FDA recommended guidelines of <3.5% RMSE. The accuracy was similar regardless of skin tone. The hospital oximeters had accuracy of 2.58% and 3.65% respectively and were less accurate in people with darker skin.

Conclusion:

The EvieMED wearable ring can accurately measure SpO₂ across the range of 70-100% oxygen and maintains consistent accuracy when evaluated on subjects with varying skin tones, likely due to its reflectance mode of measurement. An aesthetically pleasing and user-friendly wearable ring has the potential to provide longitudinal medical and wellness information to aid in the management of chronic conditions such as diabetes.

Diabetes Technology and Aviation Medicine: A Brief Review of the Current Evidence and Future Considerations

Renald Meçani, MD; Omaima El-Hakouni, BA; Monika Cigler, MD; David Russell-Jones, MD; Fariba Shojaee-Moradie, PhD; Gerd Köhler, MD; Thomas R Pieber, MD; Julia K Mader, MD; Chantal Mathieu, MD

Division of Endocrinology and Diabetology, Department of Internal Medicine, Medical University of Graz Graz, Austria

renald.mecani@medunigraz.at

Objective:

This review aims to summarize the current evidence regarding the application of diabetes management technologies in the aviation context.

Method:

A thorough literature search with MESH and non-MESH was performed on PubMed/Medline to identify studies related to diabetes management technologies in aviation.

Result:

In Europe, Austria, the UK, and Ireland permit pilots treated with insulin to operate flights under the ARA.MED.330 protocol, which requires strict and frequent capillary blood glucose monitoring. Diabetes technology is not yet approved for cockpit use. Nonetheless, recent evidence has identified CGMs as a viable alternative to CBGM. The evidence regarding insulin pumps is less favourable. In-vitro studies have demonstrated that changes in cabin pressure can lead to insulin pumps delivering excess insulin, a phenomenon attributed to the principles of fluid-gas physics Additionally, current CGMs are not approved for use above 5,500 meters in altitude, indicating that these systems might not function correctly under conditions of low air pressure, sudden decompression, or other extreme flight scenarios.

Conclusion:

The actual effectiveness of diabetes management technologies during flight is yet to be determined. Further studies in-flight and within hypobaric chambers are needed to assess the performance of these devices under various physical and physiological conditions, such as alterations in cabin pressure and different metabolic states. The results from these investigations could greatly impact the licensing procedures for diabetic pilots, potentially influencing their capability to safely regulate glucose levels while flying.

Towards Minimally Invasive Home-Glycemia Management: Development of a Glycated Albumin Electrochemical Sensor

Itsushi Minoura, PhD; Norikazu Katayama, PhD; Yuya Miyazawa, MS; Masatoshi Kayashima; Satoshi Ikeda; Yoko Kumagai, MS; Takeshi Shinohara, PhD; Koshin Sekimizu, PhD; Narushi Ito, PhD

Provigate, Inc., Bunkyo-ku, Tokyo, Japan

minoura@provigate.com

Objective:

Minimally- or non-invasive glycemia management is crucial in the management of diabetes. Serum glycated albumin (GA), the ratio of the glycated albumin to total albumin level in the serum, is a biomarker reflecting the average blood glucose levels over the previous 2‒3 weeks (HbA1c reflects the average levels over the previous 1‒2 months). It can be measured in finger-tip blood, saliva, or tear specimens. Furthermore, weekly measurement of GA has been reported to contribute to improved glycemia management. To maximize the advantage of measuring GA, utilization of a home monitoring device would be desirable. We developed a GA-measuring device that requires a minimum amount of sample.

Method:

We employed a platinum-based micro-planar amperometric hydrogen peroxide electrode overlaid with a cationselective Nafion membrane and an immobilized fructosyl amino-acid oxidase layer. The accuracy was evaluated using both fructosyl lysine (FK) solutions and protease-digested serum samples. The albumin concentration was optically measured in the presence of Bromcresol purple.

Result:

The fabricated sensor was calibrated using standard solutions, which yielded a linear current output against FK concentrations of 3‒50 μM (R²>0.999). Next, human serum was used for the evaluation; approximately 14 μL of serum was required for a single measurement. The intra-day coefficient of variation (CV) of 11 GA measurements was 1.6%, whereas that of albumin was 0.6%. Overall, the intra-day CV of GA measurements was 2.3%.

Conclusion:

We developed a sensor to measure the GA levels in serum amounts available from finger-prick blood sampling. The accuracy was sufficiently high to recommend it as a point-of-care testing device. Measurement of GA with this devicecould serve as a minimally invasive method for home glycemia management.

Evaluating Hyperglycemia Duration and Thresholds as Predictors of A1C Outcomes in Type 1 Diabetes

Eslam Montaser, PhD; Viral N. Shah, MD

IU School of Medicine, Indianapolis, Indiana, USA

emontasse@iu.edu

Objective:

Although hyperglycemia is strongly correlated with A1c levels, the precise role of glucose duration above certain thresholds remains underexplored. The objective of this study was to evaluate the impact of hyperglycemia duration and glucose thresholds (140, 180, 250 mg/dL) on A1C outcomes in individuals with type 1 diabetes (T1D) and establish baseline hyperglycemia patterns in individuals without diabetes as a reference for future interventions.

Method:

In the first phase, continuous glucose monitoring (CGM) data from 163 participants without diabetes (age 7-80 years; 66% female, 93% White) were analyzed to calculate the time above 140 mg/dL (TAR140) and categorize hyperglycemia duration into three categories (<30 min, 30-60 min, >60 min). In the second phase, we examined the relationship between A1C and hyperglycemia duration across glucose thresholds (140, 180, and 250 mg/dL) using data from 168 adults with T1D from the DCLP3 study (NCT03563313).

Result:

Among non-diabetic participants, the median daily time spent above 140 mg/dL was 25 minutes, with weak correlations to A1C, as expected. In participants with T1D, strong positive correlations were observed between A1C, and time spent above 140 mg/dL (r=0.671), 180 mg/dL (r=0.657), and 250 mg/dL (r=0.607). The strongest correlation between TAR140 and A1C occurred when hyperglycemia lasted more than 60 minutes per day. Correlations were weaker when hyperglycemia durations were shorter than 60 minutes, and TAR180 and TAR250 were not more predictive of A1C than TAR140 for durations longer than 60 minutes.

Conclusion:

A glucose threshold of 140 mg/dL, with hyperglycemia durations exceeding 60 minutes, correlated with A1C more strongly than higher thresholds or shorter durations. These findings suggest that interventions that fail to reduce postprandial hyperglycemia (TAR140) by more than 60 minutes may not significantly lower A1C, offering crucial insights for therapeutic strategies in T1D management.

Smart Offloading for Personalized Diabetic Foot Ulcer Management: Advancing Remote Patient Monitoring for Comprehensive Risk Evaluation

Bijan Najafi, PhD

UCLA, Los Angles, California, USA

najafi.bijan@gmail.com

Objective: Diabetic foot ulcers (DFUs) significantly impact patients' lives, increasing the risk of hospitalization and amputation. We propose a comprehensive remote monitoring platform to enhance DFU care.

Method: Our platform includes a standard offloading device with an integrated inertial sensor to monitor adherence and daily physical activities, a smartwatch for patient communication and data streaming to the cloud, and a remote patient monitoring portal. The portal integrates clinical information about DFUs and their severity with data from the smart offloading device, including adherence, daily steps, and mobility performance. To visualize key risk factors associated with poor healing outcomes, we developed digital health metrics scaled from 0 to 10, where higher scores indicate greater risk. These metrics provide a comprehensive overview via a visual interface. The platform captures behavioral metrics such as offloading adherence, daily steps, and cadence, combined with remotely measurable frailty and phenotypic metrics to offer an in-depth patient profile.

Results: We evaluated the feasibility of this platform with 146 DFU patients over 12 weeks; 39% experienced unfavorable outcomes, including dropout, adverse events, or non-healing. Digital biomarkers were benchmarked (010), categorized into low, medium, and high risk for unfavorable outcomes, and visually represented using colorcoded radar plots. Initial case reports demonstrate the value of this holistic visualization in identifying underlying risk factors for unfavorable outcomes, such as a high number of steps, poor adherence, and cognitive impairment.

Conclusion: While further studies are necessary to validate the effectiveness of this visualization in personalizing care and improving wound outcomes, early results in identifying risk factors for unfavorable outcomes are promising.

Hypertension and Diabetes Comorbidity: Factors That Are Associated with Their Joint Occurrence

Oluwatoyosi Ogunmuyiwa, MPH

Georgia State University, Atlanta, Georgia, United States

toyositoibat@gmail.com

Objective:

Hypertension and Diabetes are important metabolic disorders and risk factors for cardiovascular diseases. Although factors that are associated with hypertension and diabetes for cardiometabolic diseases are well established, little is known regarding factors that are associated with their joint occurrence. This study aims to determine factors that are associated with hypertension and diabetes comorbidity (HDC). This study also examines the risk factors that are independently associated with hypertension and diabetes and their joint occurrence.

Method:

This study utilized data from a cross-sectional survey of the United States Nationwide National Health and Nutrition Examination Survey (NHANES) using the 2017-2020 pre-pandemic dataset. The aim was to investigate the potential links between various demographic, behavioral, and clinical factors, and the likelihood of developing both hypertension and diabetes. Univariate and multivariate logistic regression analyses were employed to examine these associations.

Result:

An increase inage and Body Mass Index (BMI) was associated with increased odds of diabetes, hypertension, and HDC. Race/ethnicity and educational level also was found to be positively associated with increased odds of hypertension, diabetes, and HDC with men having an increased odds compared to females. NHB and Hispanics were disproportionately affected by metabolic disorders.

Conclusion:

The findings of this study suggest several factors, including age, BMI, race/ethnicity, and total cholesterol levels, were found to be closely linked to the occurrence of both diabetes and hypertension. Public health interventions should be implemented specifically for these high-risk population groups to decrease the burden of these conditions and also reduce the risk of cardiovascular disease in the United States.

Artificial Intelligence Ready and Equitable Atlas for Diabetes Insights (AI-READI): Purpose and Design

Julia Owen, PhD; Dawn Matthies, PhD; Gerald McGwin Jr., PhD; Cynthia Owsley, PhD; Sally Baxter, MD, MSc; Linda Zangwill, PhD; Cecilia S. Lee, MD, MS; Aaron Y. Lee, MD, MSCI; on behalf of the AI-READI consortium

Department of Ophthalmology, University of Washington, Seattle, Washington, USA

jpowen@uw.edu

Purpose:

The ability to understand and affect the course of complex, multi-system diseases has been limited by a lack of welldesigned, high-quality, large, ethically-sourced, and inclusive multimodal datasets. The NIH Bridge2AI program aims to remedy this shortfall in dataset availability by funding four data generation projects (DPGs) designed specifically for future AI algorithm development. The AI-READI (aireadi.org) DGP aims to generate a dataset facilitating AI discoveries in type 2 diabetes mellitus (T2DM).

Methods:

Our approach is to generate a dataset of 4000 persons >40 yrs of age that is balanced across 4 racial/ethnic groups (Asian, Black, Hispanic, White), both sexes, and 4 severities of T2DM (no diabetes, prediabetes/lifestyle-controlled, controlled by oral-medications/non-insulin injectables, insulin-dependent). Data collection sites include Birmingham AL, San Diego CA, and Seattle, WA. The variable domains of the dataset are diverse, encompassing many biomedical and behavioral aspects of health often impacted in T2DM, including retinal imaging, vision, cognition, body mass index, blood/urine testing, physical activity, EKG, continuous blood glucose, environmental exposures, social determinants of health and more. Whole genome sequencing will be performed on biospecimens. Banked blood derivatives (including serum, plasma, buffy coats, RNA, DNA and peripheral blood mononuclear cells) will be available to researchers for future proteomics, metabolomics, and other research.

Results:

The data are standardized and optimized for AI research and available to researchers through either a public-access or a controlled-access database, depending on the variables requested. The AI-READI pilot dataset of 204 individuals was released in May 2024 (https://fairhub.io/datasets/1). Data collection will occur from July 2023 through August 2026; yearly data releases are planned.

Conclusion:

This flagship dataset may enable new AI discoveries in T2DM.

Implementing a Stiffness Measurement of Red Blood Cells to Predict HbA1c

Eunyoung Park, PhD; Seungjin Kang, MS; Ung Hyun Ko, PhD

Orange Biomed, Seoul, South Korea

eunyoung.park@orangebiomed.com

Objective:

Glycated hemoglobin (HbA1c) plays a vital role as a diagnostic marker for diabetes and monitoring diabetes control. The need for a user-friendly HbA1c monitoring device to empower individuals in managing their diabetes independently is evident. Presently, accurate HbA1c tests are mainly confined to clinical labs. In this research, we devised a home-based A1c monitoring system utilizing microfluidic technology to assess the deformability of individual red blood cells (RBCs). Our study confirmed the reliability and precision of our HbA1c device, aligning with established laboratory measurement criteria.

Method:

Capillary blood samples were collected from 15 participants and more than 500 RBCs per participants were used for analysis. Blood samples were diluted in Dulbecco's Phosphate Buffered Saline solution before experimentation. The stiffness of individual RBCs was measured using our microfluidic sensor designed to mimic a blood capillary. The study adhered to the approved protocol of the Public Institutional Review Board designated by the Ministry of Health and Welfare of Korea.

Result:

RBCs from participants with higher HbA1c levels showed lower morphological elongation when passing through microchannels compared to those with lower HbA1c levels, which exhibited significant elongation in morphology. Subsequently, the distribution of RBC stiffness varied significantly among groups stratified by HbA1c levels, with those in the higher HbA1c group demonstrating a greater prevalence of stiffer RBCs compared to those in the lower HbA1c group. We confirmed a linear correlation between RBC stiffness and HbA1c levels.

Conclusion:

Our result is the first to report the proportionality between HbA1c levels and RBC stiffness. We believe that cellular stiffness measurements would be a groundbreaking metric for A1c testing, free from conventional protein quantification.

Exploiting Real-World Data and Digital Twins to Develop Effective Formulas for Dosing Insulin Boluses in Type 1 Diabetes Therapy

Elisa Pellizzari, MSc; Giacomo Cappon, PhD; Giulia Nicolis, MSc; Giovanni Sparacino, PhD; and Andrea Facchinetti, PhD

University of Padova, Padova, Italy

pellizzari@dei.unipd.it

Objective:

Developing effective meal insulin bolus (MIB) formulas based on continuous glucose monitoring sensor information is limited by the inability to exploit already acquired real-world data and test them in challenging practical scenarios. Digital twins (DTs) offer a solution to this problem, addressing the question “what would have happened if…?”. This study aims to demonstrate that the use of DTs allows exploiting the full potential of real-world data to create effective MIB dosing formulas.

Method:

The work utilized a dataset comprising 643 daily traces from the Control-To-Range 3 study, focusing on breakfast MIB. ReplayBG (Cappon et al., 2023), an open-source DT tool, was employed to identify twins of each individual and to simulate different scenarios. Specifically, we evaluated three MIB formulas: Noaro’s original model fully developed in silico (Noaro et al., 2021), Naoro’s model recalibrated via ReplayBG to the real dataset, and a newly developed XGBoost model. Evaluation included MIB estimation error vs the optimal bolus (targeted to minimize the glycemia-risk-index) and glucose control metrics.

Result:

The error in estimating the optimal bolus decreased from 4.36U with the original model to 4.12U with the XGBoost model and 4.10U with the recalibrated model. Significant improvements in glucose control were observed with both XGBoost and recalibrated models compared to the original model (time-in-range: +6.6% and +1.6%, glycemia-riskindex: -32.1% and -12.0%, respectively). XGBoost also demonstrated clinically relevant improvement in time-inrange over the recalibrated model (+5.6%).

Conclusion:

This study underscores the potential of using DTs to exploit real-world data for creating and testing new effective therapies for type 1 diabetes. By simulating different therapeutic strategies, ReplayBG enables the creation, validation and optimization of more performing MIB formulas than those developed solely in silico.

Accuracy Evaluation of Three CGM Systems against Different Comparator Devices

Stefan Pleus, PhD; Manuel Eichenlaub, PhD; Delia Waldenmaier, PhD; Stephanie Wehrstedt, PhD; Manuela Link, MD; Sükrü Öter, MD; Nina Jendrike, MD; Cornelia Haug, MD; Simon Schwaighofer, MD; Guido Freckmann, MD

Institut für Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universität Ulm, Ulm, Germany

stefan.pleus@idt-ulm.de

Objective:

The objective of this investigation was to evaluate the accuracy of the FreeStyle Libre 3 (FL3), Dexcom G7 (DG7) and Medtronic Simplera (MSP) continuous glucose monitoring (CGM) systems against different comparator devices.

Method:

24 adult participants with type 1 diabetes mellitus wore one sensor of each CGM system simultaneously over a period of 15 days. Sensors of DG7 and MSP were exchanged on days 5 and 8, respectively, while only one sensor of FL3 was used. Three, seven-hour sessions with 15-minute comparator blood glucose level measurements using YSI 2300 STAT (YSI, venous), Roche Cobas Integra 400 Plus (INT, venous) and Contour Next (CNX, capillary) were conducted on days 2, 5 and 15. Simultaneously, glucose level excursions in the hyper- and hypoglycemic range were induced. Accuracy was assessed using mean absolute relative differences (MARD) and bias as mean relative difference.

Result:

For FL3, MARDs were 11.7% against YSI (n=1929), 9.6% against INT (n=1927) and 9.7% against CNX (n=1926). Corresponding biases for FL3 were +8.5%, +3.2% and -0.9% against YSI, INT and CNX, respectively. For DG7, we found MARDs of 12.1% against YSI (n=1972), 10.0% against INT (n=1970) and 10.1% against CNX (n=1969). Analogous biases for DG7 were +7.1%, +1.9% and -2.4% against YSI, INT and CNX, respectively. Lastly, for MSP, MARDs were 11.6% against YSI (n=1859), 13.9% against INT (n=1857) and 16.6% against CNX (n=1855). Corresponding biases for MSP were -6.0%, -10.5% and -14.4% against YSI, INT and CNX, respectively.

Conclusion:

Accuracy results of all CGM systems were highly dependent on the comparator device used, which highlights the need for a standardized and traceable comparator measurement approach in CGM performance studies.

Digital Twin for Data Augmentation Enables the Development of Accurate Personalized Deep Glucose Forecasting Algorithms

Francesco Prendin, PhD; Andrea Facchinetti, PhD; Giacomo Cappon, PhD

Department of Information Engineering, University of Padova, Padova, Italy.

prendinf@dei.unipd.it

Objective:

Deep learning is increasingly being considered key within type 1 diabetes (T1D) scientific community as it enables the development of more accurate algorithms that can largely improve T1D. A common limitation of these methodologies is that their training requires extensive and diverse datasets, which are not always straightforward to collect, particularly in poor data context. To fill this gap, we demonstrated the abilities of a digital twin (DT) approach in generating extensive personalized synthetic data, thus enabling the development of more accurate deep learning algorithm for blood glucose (BG) forecasting.

Method:

We utilize and enhance ReplayBG an open-source tool for creating DT of people with T1D. Synthetic data were generated with ReplayBG to train a convolutional long short-term memory (CNN-LSTM) for glucose prediction. Personalized CNN-LSTMs were trained on a dataset from 12 patients with T1D for a training set of increasing size (from 1 to 40 days) using either the original data (ORIG) or a combination of original and synthetic data (ORIG+SYNT). Performance were evaluated in terms of root-mean-squared error (RMSE), for a prediction horizon of 30 min.

Result:

For all the training set sizes, the use of ORIG+SYNT consistently improves the RMSE of CNN-LSTM which ranges from 24.3 mg/dL to 19.1 mg/dL compared to 55.4 mg/dL to 19.7 mg/dL when using ORIG data only. Particularly, the CNN-LSTM trained on 10 days of ORIG+SYNT data provided a median RMSE=20.1 mg/dL which is comparable to the median RMSE=19.7 mg/dL provided by the CNN-LSTM trained on 40 days of ORIG data only.

Conclusion:

Leveraging ReplayBG for generating personalized synthetic data has the potential to mitigate data scarcity, enabling the development of more accurate deep learning algorithms.

Accounting for Hypoglycemia Treatments in Continuous Glucose Metrics

Elliott C. Pryor, BS; Anas El Fathi, PhD; Marc D. Breton, PhD

University of Virginia, Charlottesville, VA, United States

hyy8sc@virginia.edu

Objective:

Continuous glucose monitoring (CGM) is increasingly used in the management of type 1 diabetes (T1D), empowering technologies like automated insulin delivery and decision support. However, behavioral factors like hypoglycemia treatments (HT) are not captured in CGM data. HTs reduce the frequency and duration of hypoglycemia events recorded by CGM, biasing CGM-based outcomes (e.g. time below range) and potentially leading to suboptimal decision-making by systems relying on CGM data. To overcome this, we propose a method to incorporate HTs directly into hypoglycemia metrics.

Method:

We hypothesize that hypoglycemia reduction depends on treatment proactiveness and potential severity. We fit a function of (i) glucose at treatment and (ii) projected minimum glucose without treatment to the amount of avoided hypoglycemia. Additionally, we introduce an HT detector using CGM data to identify HTs without explicit labels.

Result:

Our function has an R² of 0.94 on in silico data comparing TBR with and without HT, and our HT detector has an F1 score of 0.72 on clinical data with labeled HT. We demonstrate an application of this methodology in a published run-to-run adaptation system using CGM-based metrics. In silico we reduce the average number of HT per day from 3.3 in the original system to 1.6 with the modified metrics, while maintaining 84% time in 70-180 mg/dL.

Conclusion:

Our proposed method ensures a comprehensive analysis of CGM data. Our results indicate that this function can be seamlessly integrated with existing CGM-based methods in the literature, allowing for more robust management of T1D by accounting for the variability in HT behaviors within hypoglycemia metrics.

Non-Invasive Glucose Measurement Without Per-Person Calibration Using Raman Spectroscopy

Fabien Rebeaud, PhD; Marc Stoffel, MD; J. Hans DeVries, MD, PhD; Ismene Grohmann MSc; Adler Perotte, MD, MA

Liom (Spiden AG), Pfaeffikon SZ, Switzerland

fr@liom.com

Objective:

We investigated the accuracy of a novel Raman spectroscopy-based system and associated computational model for the non-invasive, continuous measurement of glucose.

Method:

This is a first analysis of a single-centre, multiple-cohort open clinical study (NCT06272136). Laser-based photonics technology was used to collect Raman spectra continuously at the wrist while adults living with type 1 diabetes mellitus (DM) underwent a meal challenge with delayed insulin bolus administration. We developed and tested a computational model to infer glucose concentration from the acquired spectra. Noninvasively collected glucose values were compared with reference glucose values.

Result:

This cohort comprised 14 adults with type 1 DM (6 men and 8 women, aged 26 to 63, Fitzpatrick skin types 1 to 4) 857 paired data points were collected at 5-minute intervals. Measured glucose reference values ranged from 62 to 393 mg/dl. Without per-subject calibration (“calibration-free”), we report a mean absolute relative difference (MARD) of 20.5%, R² of 0.71, and 96.3% of paired data points in the A + B region of a Parkes Consensus Error Grid. With persubject calibration, we found a MARD of 7.3% (R², 0.96) and 99.8% of paired data points in the A + B region of the error grid. No device-related adverse events were seen.

Conclusion:

Our findings suggest that measuring glucose noninvasively without calibration is possible and safe. Future work will aim to apply advanced machine learning models to improve the calibration-free MARD and apply this model to a larger and more diverse cohort.

Real World Efficacy of the iLet Bionic Pancreas

Steven J. Russell, MD, PhD; Rajendranath R. Selagamsetty, MS; Edward R. Damiano, PhD

Beta Bionics and Massachusetts General Hospital, Concord, MA, USA and Boston, MA, USA

srussell@betabionics.com

Objective:

To analyze the impact of the iLet bionic pancreas on glycemic control during the first year after FDA clearance and compare to the results of the Bionic Pancreas Pivotal Trial (BPPT).

Method:

Commercial users of the iLet who had a pre-iLet HbA1c value available and had at least 3 weeks of iLet data in the cloud were included. Baseline HbA1c values were compared with the glucose management indicator (GMI) values calculated from all available continuous glucose monitoring (CGM) data, both reported as mean±SD. Adults (≥18 years) and children were analyzed separately. Data from the BPPT were analyzed in the same way for comparison. The percentage of time with CGM glucose <54 mg/dl during iLet us was calculated for all groups and reported as median(IQR).

Result:

Data from 2,965 adults and 540 children were included in the real-world analysis and from 218 adults and 112 children in the BPPT analysis. Adult commercial users had a baseline HbA1c of 8.4±1.9% and an iLet GMI of 7.2±0.4% (difference:–1.2%), compared to 7.7±1.2% and 7.0±0.3%, respectively, in the BPPT (difference:–0.7%). Child commercial users had a baseline HbA1c of 9.1±2.1% and an iLet GMI of 7.7±0.5% (difference:–1.4%), compared to 8.1±1.2% and 7.4±0.3%, respectively, in the BPPT (difference:–0.7%). The time with CGM glucose <54 mg/dl was 0.3(0.1-0.6)% in adults and 0.3(0.1–0.6)% children in commercial use and 0.3(0.1–0.5)% in adults and 0.3(0.2–0.6)% in children in the BPPT.

Conclusion:

The baseline HbA1c values of commercial users were higher, and the decreases from baseline HbA1c to iLet GMI values were larger than among participants in the BPPT. The time spent <54 mg/dl were comparable in both settings.

Unlocking Potential: Personalized Lifestyle Therapy for Type 2 Diabetes Through a Predictive Algorithm-Driven Digital Therapeutic

Swantje Kannenberg, MD; Jenny Voggel, PhD; Nils Thieme, MS; Oliver Witt, PhD; Kim Lina Pethahn, MS; Morten Schütt, MD; Christian Sina, MD; Guido Freckmann, MD; and Torsten Schröder, MD, PhD

Torsten Schröder, MD, PhD, Research & Development, Perfood GmbH, Lübeck, Schleswig-Holstein, Germany.

torsten.schroeder@perfood.de

Background:

We present a digital therapeutic (DTx) using continuous glucose monitoring (CGM) and an advanced artificial intelligence (AI) algorithm to digitally personalize lifestyle interventions for people with type 2 diabetes (T2D).

Method:

A study of 118 participants with non–insulin-treated T2D (HbA1c ≥ 6.5%) who were already receiving standard care and had a mean baseline (BL) HbA1c of 7.46% (0.93) used the DTx for three months to evaluate clinical endpoints, such as HbA1c, body weight, quality of life and app usage, for a pre-post comparison. The study also included an assessment of initial long-term data from a second use of the DTx.

Results:

After three months of using the DTx, there was an improvement of 0.67% HbA1c in the complete cohort and −1.08% HbA1c in patients with poorly controlled diabetes (BL-HbA1c ≥ 7.0%) compared with standard of care (P < .001). The number of patients within the therapeutic target range (< 7.0%) increased from 38% to 60%, and 33% were on the way to remission (< 6.5%). Patients who used the DTx a second time experienced a reduction of −0.76% in their HbA1c levels and a mean weight loss of −6.84 kg after six months (P < .001) compared with BL.

Conclusions:

These results indicate that the DTx has clinically relevant effects on glycemic control and weight reduction for patients with both well and poorly controlled diabetes, whether through single or repeated usage. It is a noteworthy improvement in T2D management, offering a non-pharmacological, fully digital solution that integrates biofeedback through CGM and an advanced AI algorithm.

Integrating Continuous Glucose Monitoring Data into Electronic Health Records: A Solution Conforming to iCoDE 2022 Standards

Conor Sheehy, Pharm.D,; Markus Bertolozzi; Anthony DiPietro, BCS

TheraponHealth, Baltimore, Maryland, USA

conor@theraponhealth.com

Objective:

Integrating continuous glucose monitoring (CGM) data into electronic health records (EHR) enhances diabetes management by offering real-time insights into patient glycemic patterns. This project outlines the development of a CGM-EHR integration solution adhering to the 2022 International Consensus on Diabetes and Endocrinology (iCoDE) standards.

Method:

Our solution is an EHR agnostic cloud-based CGM as a Service (CGMaaS) platform that securely integrates CGM data into major EHR systems. Built on Platform as a Service (PaaS) architecture, it utilizes Infrastructure as a Service (IaaS), Data as a Service (DaaS) via Snowflake, and the Software as a Service (SaaS) technologies Tableau for data visualization. We utilize Advanced Application Programming Interfaces (APIs) and Continuous Integration/Continuous Deployment (CI/CD) pipelines to ensure seamless data intake, storage, and analysis.

Result:

The solution successfully integrated patient data into Epic EHR through FHIR APIs, enabling real-time CGM updates and bidirectional communication. The end user can access glucose reports, trends, and alerts via a userfriendly Tableau interface, ensuring compliance with iCoDE standards for privacy and data accuracy.

Conclusion:

The integration of CGM data into Epic EHR using FHIR APIs demonstrates Apollo’s potential to improve diabetes management in various clinical settings. Adherence to iCoDE standards ensures scalability, security, and interoperability. Future enhancements will expand integration with additional EHRs and refine user interfaces to further optimize clinical impact.

Intradermal Glucose and Lactate Dual Sensor Feasibility for Integrated Diabetes Management

Cristian Silva, PhD, Bo Wang, PhD, Liang Wang, PhD, Rebecca Gottlieb, PhD, Naresh Bhavaraju, PhD., Iñigo San Milán, PhD., Jared Tangney, PhD

Biolinq, San Diego, California, USA

cristian@biolinq.com

Objective:

Automated Insulin Delivery (AID) systems have revolutionized diabetes management, showing safety and effectiveness in improving glucose control. However, they still require manual input to inform the algorithm of exercise to avoid hypoglycemia. Lactate is an important biomarker of exercise exertion and can be measured continuously in interstitial fluid to provide real-time input to AID systems on exercise intensity. To increase safety and reduce user burden, Biolinq has created a continuous intradermal sensor that measures both lactate and glucose and contains an embedded accelerometer to measure steps. This study evaluated the feasibility of the intradermal sensor in both in-clinic and real-world environments.

Method:

Dual glucose-lactate sensors (n=19) and lactate-only sensors (n=16) were placed on the forearms of 9 healthy subjects. They performed an exercise challenge (stationary bike with increasing resistance) with capillary samples every 3-5 minutes. To gather real-world extreme use data, a lactate sensor was placed on 23 professional soccer players (‘athletes’). The athletes performed a treadmill assessment with increasing speed and/or field training. A retrospective algorithm was used to correlate capillary reference data from Roche Accu-Chek Guide Me (glucose) or the Nova Biomedical, Lactate Plus (lactate).

Result:

The healthy subjects show 54.9% of lactate channels within 20% and 86.6% within 40% of reference lactate samples (n=749, Fig 1). Glucose channels showed 99.3% values within 20% of reference with a MARD of 5.9% (276 samples). In the athletes, the intradermal sensors tracked lactate both in treadmill and field testing.

Conclusion:

Continuous lactate and glucose monitoring is feasible on a single wearable intradermal sensor and, with further development, can provide real-time input to inform AID systems on exercise intensity.

Exploring the Utility and Acceptance of Continuous Glucose Monitor (CGM) Use Among Hispanic Patients with Insulin-Treated Type 2 Diabetes: A Pilot Focus Group Study

Diana Soliman, MD, MHS; Giuliana Arevalo, MD; Rodolfo J. Galindo, MD; Daniel E. Jimenez, PhD; Francisco J. Pasquel, MD, MHS; David Baidal, MD; Ashby F.Walker, PhD

Division of Endocrinology, Diabetes and Metabolism, Department of Medicine, University of Miami Miller School of Medicine, Miami, FL, USA

dsoliman@med.miami.edu

Objective:

Hispanic individuals are disproportionately affected by type 2 diabetes (T2D) and face disparities in diabetes technology utilization. Despite proven benefits, continuous glucose monitoring (CGM) use remains low among Hispanic adults with insulin-treated T2D. This study used a social ecological model framework to identify individual, cultural, and societal-level barriers to CGM access and adoption among this population.

Method:

Two sequential focus groups were conducted in Spanish at the University of Miami with Hispanic adults (n=11) with T2D. Inclusion criteria were HbA1c ≥8%, ≥1 daily insulin injection, and no CGM use in the past 2 years. The initial focus group was conducted to identify barriers to CGM initiation. After the focus group, participants were provided with DexCom G7 CGMs for 30 days of use. After completing the wear period, the same cohort of participants returned for a follow-up focus group, aimed to explore factors related to CGM adoption and acceptance. Focus groups were analyzed using a thematic analysis, by systematically coding and categorizing key concepts and ideas expressed by participants.

Result:

The mean age was 53.09 ± 10.20 years and baseline HbA1c 9.82 ± 1.44%. Individual-level barriers included limited awareness of CGM existence, concerns about device functionality and comfort, and poor self-advocacy skills. Cultural barriers were limited but having Hispanic/Latino healthcare providers with shared cultural understanding was identified as a facilitator for overall diabetes management. Societal-level barriers encompassed high costs, insurance limitations, and insufficient information exchange during provider visits.

Conclusion:

This pilot study provides novel insights into the experiences of the Hispanic population and highlights the complex interplay of individual, cultural, and societal factors influencing CGM use. Tailored interventions addressing these barriers are crucial for improving CGM utilization in this population.

Examining Sex Differences in Glycemic Outcomes Among Adults Enrolled in an Inpatient Continuous Glucose Monitoring as Standard of Care (CGM as SOC) Program

Samantha R. Spierling Bagsic, PhD, MSE; Addie L. Fortmann, PhD; Rebekah Belasco, MS; Haley Sandoval, BA; Alessandra Bastian, RD, MBA, MPM; Suzanne Lohnes, FNP, MSN; Laura Talavera, MSN; Emily C. Soriano, PhD; Athena Philis-Tsimikas, MD

Scripps Health, San Diego, CA, USA

Bagsic.samantha@scrippshealth.org

Objective:

Given known sex and gender differences in diabetes risk factors and outcomes—yet the paucity of glucometric data in the hospital setting—a large dataset stemming from a COVID-era CGM as SOC program with over 1,000 enrollments serves a novel opportunity to investigate sex-differences in glycemic control and in the demographic and clinical characteristics known to commonly impact glycemic control in the hospital setting.

Methods:

Hospitalized adults with diabetes and subcutaneous insulin orders were enrolled from May 2020 – June 2023. EHR data were obtained using the Snowflake cloud environment and integrated with CGM data to form a dataset including clinical and demographic characteristics present upon or within 24 hours of admission, hospital outcomes, and point-of-care glucose and CGM data during the hospital stay.

Results:

More men (n=572, 56%) than women (n=442, 44%) were enrolled in the CGM as SOC program. Men were more likely to smoke (58% vs 36%), while more women used anti-hyperglycemic medications pre-hospitalization (70% vs 64%), were older (67 vs 65 years), and had higher BMI (32.3 ± 9.4 vs 30.1 ± 8.2 kg/m²; ps < .05). Men had higher mean HbA1c (8.6 ± 2.5 vs 8.2 ± 2.4%) than women and reached higher maximum POCT glucose readings upon admission (279 ± 138 vs 262 ± 133 mg/dL; ps < .05). After adjusting for these baseline differences, no sex differences were observed on any CGM metrics measured, including mean glucose, COV, % TIR, TBR, or TAR.

Conclusions:

In a large real-world evaluation of CGM as SOC in the hospital, despite baseline sex differences in demographic and clinical characteristics, glycemic control measured by CGM was comparable for males and females.

How Sweet It Is: Implementing Continuous Glucose Monitoring in Gestational Diabetes

Rachel Stahl Salzman, MS, RD, CDN, CDCES; Jane Jeffrie Seley, DNP MSN MPH GNP BC-ADM CDCES; Felicia A. Mendelsohn Curanaj, MD

Division of Endocrinology, Diabetes and Metabolism, Weill Cornell Medicine, New York, New York, United States

esr9014@med.cornell.edu

Objective:

Gestational Diabetes (GDM) is on the rise in the U.S. and worldwide. GDM contributes to adverse maternal and fetal outcomes, doubles the risk of maternal type 2 diabetes, and increases risk of obesity and diabetes in offspring. FDA-approval of continuous glucose monitoring (CGM) in GDM offers the opportunity to wear CGM as an alternative to blood glucose monitoring (BGM).

Method:

The Division of Endocrinology, Diabetes and Metabolism in collaboration with Department of Obstetrics and Gynecology (OB/GYN), launched an implementation project consisting of a Virtual GDM Program with a virtual class on lifestyle modification and glucose monitoring, offering the choice of BGM or CGM or a combination. A second class on safe and effective use of CGM in GDM was offered, and a visit with a Registered Dietitian/Certified Diabetes Care and Education Specialist and endocrinologist, if needed. REDCap data collection included demographics, class attendance, CGM metrics and barriers to wearing CGM. CGM satisfaction was evaluated using a Modified Glucose Monitoring Satisfaction Survey (GMSS) via Qualtrics.

Result:

We found high acceptance of CGM, with 58% choosing CGM over BGM. The majority of those wearing CGM (73%) wore it for > 70% of the time over 14 consecutive days. All participants rated our virtual classes positively and would recommend them to others. Results from the Modified GMSS showed high satisfaction with CGM, especially understanding the impact of food & physical activity on glucose levels.

Conclusion:

Our implementation project showed the feasibility of implementing CGM for people with GDM using a standardized protocol. Further research is needed to enhance knowledge and comfort level of using CGM data by OB/GYN and establishing CGM metrics in individuals with GDM.

In Vivo Evaluation of Novel Long-Term Intravascular Continuous Blood Glucose Monitor in a Chronic Ovine Model

Mark A. Tapsak, PhD; Paul V. Goode, PhD; JP Thrower, PhD; Michael Talcott, D.V.M.; Timothy L. Routh, M. Eng.; Stephen T. Tapsak; Jose Garcia; Samantha Wakil, PhD

Glucotrack Inc., Front Royal, VA, USA

mtapsak@glucotrack.com

Objective:

This study provides in vivo assessment of Glucotrack's long-term implantable Continuous Blood Glucose Monitor (CBGM) in an ovine model. The primary objectives were to evaluate intra- and inter-device performance, assess the suitability of sensing in the vascular space, and examine chronic biocompatibility and histology through explant and necropsy.

Method:

Across 4 studies, 17 male sheep each received up to 2 intravascular device implants—one in the left and one in right jugular vein. Devices were explanted at either 30, 60, or 90 days post-implantation to generate histopathological data in addition to the continuous sensor data. Blood glucose measurements were performed regularly. Glucose excursions were conducted periodically during the study and compared against conventional blood glucose monitoring values measured in triplicate. MARD (CBGM-BGM) and paired absolute relative difference (PARD) (CBGM-CBGM) scores were calculated during glucose challenge tests throughout the respective implantation periods. Comprehensive blood analyses were performed at the beginning and end of the study.

Result:

The results demonstrated successful intravascular implantation (~20 minutes) without sophisticated techniques or tools. During the study, no serious adverse events were observed. Notably, MARD improved over time, indicating sustained accuracy of the Glucotrack CBGM. MARD across the 33 devices was regularly <8%, with PARD often showing even better results.

Conclusion:

The stability observed over 90 days in a large and diverse sample, coupled with the device being well-tolerated within the blood vessel, suggests that Glucotrack’s CBGM could be a reliable and accurate long-term solution for individuals managing diabetes without the need for an external wearable.

Effectiveness of a Telehealth-Based Digital Therapeutics Program on Glycemic Control in Type 2 Diabetes Mellitus (T2DM): A Quasi-Experimental Study

Premalatha Thiyagarajah, MS; Tarek Turk, MBBS, MD, PhD; Shraddha Kavugoli, MS

Trudoc Healthcare LLC, Mumbai, Maharashtra, India

drpthiyagarajah@trudochealth.com

Introduction:

Diabetes is a leading cause of morbidity and mortality globally, affecting individuals across all demographics and imposing a high social and economic burden. The global age-standardized prevalence of diabetes is 6.1%.[1]Effective diabetes management requires continuous monitoring, highlighting the need for digital health systems. This study evaluates impact of a telehealth program on diabetes management via the Wellthy Care Digital Therapeutics Platform.The integration of telehealth in diabetes care has proven effective in providing personalized, real-time support to patients, enhancing their ability to manage the condition effectively.

Method:

This quasi-experimental pre-post uncontrolled study included 159 adults with T2DM (47.2% female, 52.8% male) with a mean age of 44.71 years, participating in a 40-week telehealth program through exercise plans, medication adherence, risk management, self-monitoring, and stress management to improve diabetes outcomes. Statistical analysis using SPSS 21 included descriptive analysis and paired t-tests, with p < 0.05 considered significant.

Results:

A paired t-test showed a significant reduction in HbA1c levels from baseline (9.86 ± 1.91) to the endline (7.69 ± 1.61), a decrease of 2.17 (95% CI, 1.99 to 2.35), t(158) = 23.772, p < 0.001, with a strong positive correlation (r = 0.799, p < 0.001), indicating the intervention's effectiveness. The significant reduction in HbA1c levels from baseline to end of the study underscores the potential of digital therapeutics to enhance diabetes management.

Conclusion:

The study found that the Digital Therapeutics Platform significantly improved glycemic control in patients with type 2 diabetes mellitus. The significant reduction in HbA1c levels from baseline to the end of the study underscores the potential of digital therapeutics to enhance diabetes management.Furthermore, the incorporation of digital health tools can lead to more efficient use of healthcare resources and better patient engagement. These results suggest that widespread adoption of such platforms could substantially alleviate the global burden of diabetes.

Pharmacokinetic Properties of Once-Weekly Insulin Icodec in Children and Adolescents with Type 2 Diabetes

Michelle Van Name, MD; Björg Ásbjörnsdóttir, MD, PhD; Mikel Murphy Gomes, MSc; Paula M. Hale, MD; Keerthana Udupa, MSc; Siri Vinther, MD, PhD; Rasmus Ribel-Madsen, PhD

Yale School of Medicine, Pediatric Endocrinology & Diabetes, New Haven, CT, USA

michelle.vanname@yale.edu

Objective:

Glycemia is above target in many youths with type 2 diabetes (T2D). Insulin icodec is a novel basal insulin engineered for once-weekly administration and holds potential to ease the treatment burden of daily injections. The efficacy and safety of once-weekly icodec has been demonstrated in a phase 3a program in adults with diabetes. The objective of this study was to investigate the pharmacokinetic properties of icodec in children and adolescents with T2D.

Method:

In a one-period, open-label study, 18 insulin-treated children and adolescents with T2D aged 10 to <18 years (females/males 11/7; mean±SD BMI 38.5±7.1 kg/m², HbA_1c 7.5±1.3%) received a single subcutaneous icodec administration (5.6 U/kg). Blood was sampled for pharmacokinetics until 35 days post-dose. Observed single-dose pharmacokinetic profiles were extrapolated to steady state using pharmacokinetic modelling.

Result:

The shapes of the observed single-dose and modelled steady-state pharmacokinetic profiles in children and adolescents were comparable to those previously seen in adults with T2D. After a single dose, geometric mean total exposure was 81,901,778 pmol∙h/L (CV 28%) and maximum concentration was 421,975 pmol/L (CV 26%). Median time to maximum concentration was 21 hours. When extrapolated to steady state, total exposure was 77,100,927 pmol∙h/L (CV 30%) and maximum concentration was 635,464 pmol/L (CV 27%). A total of 6 adverse events (AEs) in 2 participants were reported (all mild and recovered/resolved). No serious or severe AEs, no hypersensitivity or injection site reactions, no hyperglycemic episodes and one level 1 hypoglycemic episode (plasma glucose <70 mg/dL and ≥54 mg/dL) were reported.

Conclusion:

Icodec was well tolerated in the current study and the pharmacokinetic properties support the potential of icodec for once-weekly administration in children and adolescents with T2D.

Using Commercial Artificial Intelligence to Assess Nutrients from Photographs of Meals of Type 2 Diabetes Patients

Kayo Waki, MD, MPH, PhD; Daniel R. Lane, MS, MBA; Yuexiang JI,

The University of Tokyo, Bunkyo-ku, Tokyo, Japan

kwaki-tky@m.u-tokyo.ac.jp

Objective:

Improving diet is key in type 2 diabetes treatment. Interventions often have poor adherence due to burdensome food logging methods. Approaches using food photographs assessed by artificial intelligence (AI) may make food logging easier, if they are adequately accurate. Large language model models such as generative pre-trained Transformer (GPT) are making rapid advances. Our objective was to determine the usability and accuracy of a current commercially-available frontier GPT for the task of identifying the nutritional content of a meal using as input a photograph of the meal.

Method:

From previous work, we had data for 22 home-cooked Japanese meals, including weighed food records (providing nutritional ground truth), photographs, and nutritional estimates provided by dieticians using only these photographs. We used OpenAI’s gpt-4o model (gpt-4o-1014-05-13) with one-shot prompts and no fine-tuning to assess energy, fat, protein, carbohydrate, fiber, and salt of photographs of the 22 meals, comparing assessments to the weighed food records for each meal and to the assessments of dieticians.

Result:

The model had poor performance overall. Results for energy, fat, protein, carbohydrates, and salt had very low intraclass correlation coefficients (ICCs) ranging from 0.04 and 0.36, well below the moderate-to-good 0.46 to 0.74 ICCs achieved by the dieticians. For fiber, though, the model achieved a fairly high ICC of 0.71 (0.67-0.74 95% CI), well above the dietician fiber performance of 0.57.

Conclusion:

Current AI accurately assesses fiber content in meals but is inaccurate for other nutritional parameters. The good accuracy with fiber makes this method a viable approach to log meals and provide reasonable accurate dietary feedback to patients as part of an intervention to increase fiber intake.

Dual Sensing Glucose-Insulin Microneedle

Braeden Weems; Sophia Edwards; Isaac Helmer; Christopher Correa; Logan Keller; Annika Banuelos, BS; Miguel Fermaint, MS; Taliah Gorman, MS; Koji Sode, PhD; Jeffrey T. La Belle, PhD

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

Braeden.Weems@gcu.edu

Objective:

Monitoring levels of glucose and insulin in a single sensor can aid in the reduction of hypoglycemia by allowing proper insulin dosing at specific levels of glucose. Opting to place both in one continuously monitoring sensor targeting interstitial fluid opposed to capillary blood sampling, can reduce pain while minimizing damages to skin. To accomplish this, we propose a microneedle array design capable of reading varying glucose-insulin concentrations using insulin antibody and glucose dehydrogenase (GDH) immobilized in a hydrogel layer.

Method:

Electrochemical Impedance Spectroscopy (EIS) and Cyclic Voltammetry (CV) were performed on a gold-sputtered microneedle array made of UV curable resin. Ruthenium-Poly(ethylene glycol) diacrylate (RuT-PEGDA) was spun onto one array to immobilize GDH. The imaginary impedance response and optimal frequency of glucose was taken in a bath of Phosphate-buffered saline, 100 mM K-Ferricyanide, and glucose.

Result:

Preliminary testing for glucose was complete using EIS and CV. The Formal Potential (FP) of the arrays with GDH immobilized RuT-PEGDA were calculated to be0.0882 V. The imaginary impedance had an R²value of 0.8508 at a 0.0020 slope with the frequency optimized to 250.9 Hz. Insulin data collection is in progress.

Conclusion:

Here we tested individual immobilizations of glucose. The next steps are to complete insulin testing and move on to co-immobilization, verify reproducibility, run co-immobilized samples through flow cells at physiological conditions mimicking interstitial fluid, and substitute the K-Ferricyanide redox probe with a biocompatible one in preparation for animal studies.

Machine-Learning Predictions of Daily Glucose Fluctuations Using Scalp Electroencephalogram (EEG)

Katherine Werbaneth, Teresa Wu, Tina Denison, Heonsoo Lee, Emily Mirro, Casey Halpern, Irina Nayberg, Joann Angeles, David Klonoff

Introduction:

The brain is known to be the primary regulator of the body’s homeostasis and play a pivotal role in orchestrating glycemic control. We previously found that the brain’s diffuse electrical rhythms can be decoded into quantifiable projections of glucose fluctuations. Due to this well-distributed nature of brain rhythms correlating with glucose swings, we hypothesized that scalp-mounted electroencephalogram (EEG) electrodes could detect these metabolic signals and give access to our novel machine learning algorithms to predict daily glucose fluctuations for people in need of better metabolic control.

Methods:

We conducted two experiments to examine scalp EEG signal changes in relationship to blood glucose fluctuations. The first (n=16) was conducted in a hospital environment over 2-9 consecutive days where we collected standard scalp EEG data and continuous glucose levels via a continuous glucose monitor (CGM, Dexcom G6 Pro). This study environment allowed us to capture sleep and wake periods, including meals. The second study (n=18) was to examine the glycemic response to one-time glucose drink (75mg; Azur Medical). During the experiments, interstitial (Experiment 1) and blood glucose (Experiment 2) were continuously collected along with scalp EEG.

Results:

In both experiments a novel machine learning framework was applied, and entropy and spectral features of EEG were used in the model. The EEG-predicted glucose levels exhibited correlation (r=0.56) to the interstitial/blood glucose levels in both long (continuous days including sleep/wake cycle and meals) and short (response to fixed amount of glucose) time scales. Furthermore, maximum glucose from CGM over 1-4 days and that of our prediction revealed a strong agreement (r=0.88, p<0.0001).

Conclusion:

Our results demonstrate the potential of a completely non-invasive modality (i.e. EEG) as a method for long-term monitoring of glucose fluctuations.

Quantifying Hypoglycemia Risk Across eHbA1c Levels: Suggesting a Safe Lower Limit for People with Diabetes

Andrew Hudson Yang; Vincent Feng Yang, MSc

University of California San Diego, La Jolla, California, United States of America

any012@ucsd.edu

Objective:

This study investigates the relationship between hemoglobin A1c (eHbA1c) levels and hypoglycemia incidence in people with diabetes, aiming to propose a justified lower boundary for the recommended eHbA1c in glycemic control targets.

Method:

An analysis was conducted on historical Continuous Glucose Monitoring System (CGMS) data from 652 people with diabetes. eHbA1c values were correlated with the frequency and duration of hypoglycemic events.

Hypoglycemic targets are defined as maintaining hypoglycemia (<3.9 mmol/L) below 4% and severe hypoglycemia (<3.0 mmol/L) below 1%. The proportion of individuals experiencing hypoglycemia and the average hypoglycemia duration across different eHbA1c values were assessed.

Result:

Our analysis revealed a significant increase in those failing to meet the hypoglycemia target (<3.9 mmol/L) as eHbA1c decreased. Specifically, failure rates rose from 35.56% in those with eHbA1c below 6% to 80.52% with eHbA1c below 5%. All individuals with eHbA1c below 4.75% failed to meet the target, highlighting the critical risk of excessively low eHbA1c levels. Each 0.5% decrease in eHbA1c from 6% was associated with 50% increased risk of target failure. Conversely, the proportion of individuals meeting the target declined from 92.09% above 6% eHbA1c to 76.23% above 4% eHbA1c, with a 4.6% decrease in success likelihood per 0.5% drop in eHbA1c. These results suggest the lower boundary for recommended eHbA1c should be set well above 4.75% and below the consensus upper limit of 6.5% for eHbA1c control. The possible suggested lower boundary is between 5.0% and 5.5%.

Conclusion:

We identified a strong inverse relationship between eHbA1c levels and hypoglycemia risk. We recommend setting the lower boundary for eHbA1c at 5.5%, as it optimally balances the risk of not meeting the hypoglycemia target.

Anti Diabetic Activity of Undaria Pinnatifida and Moringa Oleifera

Sai Kalyani Yogini C, MSc, PhD

Department of Bio Medical Sciences, School of Health Sciences, The Apollo University, Chittoor. Andhra Pradesh, India

saikalyani_y@apollouniversity.edu.in

Objective:

The aim of this work was to evaluate the inhibitory activities of ethanolic extracts of Undaria pinnatifida and methanolic extract of Moringa oleifera and the combination of both the extracts in 1:1 ratio.

Method:

Alpha amylase and alpha glucosidase inhibitors are used to achieve greater control over hyperglycaemia in type 2 diabetes mellitus. The present study intends to screen novel alpha amylase and alpha glucosidase inhibitors from natural sources like plants to minimize the toxicity and side effects of the inhibitors currently used to control hyperglycaemia.

Result:

The alpha amylase inhibition assay showed best efficacy in combination extract UPEA and MOM in (1:1 ratio) (90.34%) than UPEA (85.49%) and MOM (66.62%) individually exhibited at 200 µg/mL concentration. The alpha glucosidase assay showed best efficacy in combination extract UPEA and MOM in (1:1 ratio) (83.81%) than UPEA (69.34%) and MOM (64.64%) individually exhibited at 200 µg/mL concentration.

Conclusion:

The results of the work therefore clearly indicate the potential of these individual extracts exceeds by combination extract to manage hyperglycaemia.

Point-of-Care A1c Screening in the Emergency Department

Anum Zehra, BA; Rana Malek, MD; Elizabeth Fitch, RN; Kashif M. Munir, MD

University of Maryland School of Medicine, Baltimore, MD, USA

Anum.zehra@som.umaryland.edu

Objective:

Many US adults with diabetes and most with prediabetes are undiagnosed. In inner-city populations, many individuals lack a primary care provider (PCP) and seek healthcare in the emergency department (ED). Targeted real-world screening approaches may help identify individuals at risk. We sought to assess the prevalence of undiagnosed prediabetes and diabetes by testing Hemoglobin A1c (HbA1c) of adults presenting to the University of Maryland ED.

Methods:

All individuals presenting to the ED who were ≥ 18 years old and had an incidental serum blood glucose of ≥100 mg/dL were approached for point-of-care HbA1c testing. Those with a known history of prediabetes or diabetes, and for whom HbA1c data was missing were excluded.

Results:

Out of 147 individuals without a known history of diabetes or prediabetes, 45 declined testing and data was unavailable for 7 people for “other” reasons. 48 of 95 tested individuals (50.5%) had HbA1c between 5.7% and 6.4%, and 10 (10.5%) had an HbA1c ≥ 6.5%. Higher A1c subgroups had significantly more individuals with BMI > 25.0 kg/m2 (p value=0.004) and BMI >29.9 kg/m2 (p value=0.048). 100% and 60% of the 10 individuals with HbA1c ≥ 6.5% had a BMI > 25.0 kg/m2 and BMI > 29.9 kg/m2, respectively. Although not statistically significant, linkage with a PCP was lowest in HbA1c > 6.5% subgroup (p value=0.064)

Conclusion:

Screening for prediabetes and diabetes in an ED may help reduce the number of undiagnosed individuals. Using a targeted real-world screening approach (glucose ≥ 100 mg/dl) provides decreased resource utilization compared with universal screening. Defining additional variables (e.g. BMI) associated with diagnosis may further refine the screening approach to maximize case detection while limiting resources.