Abstract
DTM 2023 Abstracts
Continuous Glucose Monitoring Among Patients with Type 1 Diabetes in Rwanda: Update from the CAPT1D Study
Jason Baker, MD; Jean-Claude Habineza, MD; Giacomo Cappon, PhD; Christian Schütz, MD, PhD; Etienne Uwingabire, MD; Paul Mbonyi, MD; Steve Mey, MSc; Sandhya Narayanan, EdD, RDN, CDCES; Alvera Muka, MD; Diana L. Malkin-Washeim, PhD, MPH, RDN, CDCES, CD-N
Diabetes Empowerment International New York, NY, USA
Continuous glucose monitoring systems (CGM) have shown clinical significance by increasing time in glucose range and quality of life among diverse populations. CGM is limited in underserved communities, globally. The aim of the CAPT1D study is to demonstrate feasibility of CGM among patients with T1D in Rwanda, including assessment of time in range, HbA1c and rates of hypoglycemia, hyperglycemia, and diabetes-related hospitalization.
Fifty participants with T1D were recruited (27.24 ± 3.59 years, 40% male) in a single-arm 12-month prospective study. Glucose management was evaluated with HbA1c tests and CGM data. Four questionnaires were used to collect information in regard to diabetes related hospitalizations, hypoglycemia confidence, diabetes self-management and any adverse events at regular intervals.
Results reported here are for HbA1c and CGM data up to 9 months. CGM data were analyzed using Automated Glucose dATa Analysis (AGATA), an open-source software. In compliance with current international consensus definitions, mean glucose level, time in range and the glycemic risk index (GRI) were compared at 9-months versus the baseline.
A significant improvement in glycemic control at 9 months (p < 0.01) was found. HbA1c levels decreased by 2.81%, and time in target and tight target ranges increased by 12.91% and 9.28%, respectively, with a reduction of time in hyperglycemia of 13.03%. Increase of time in hypoglycemia was significant but minimal (0.12%). Overall mean glucose decreased (-38.50 mg/dl), and glycemic risk dropped -12.95).
The trending data from the ongoing CAPT1D study demonstrates the feasibility of implementing CGM technology in an underserved population in Rwanda. The findings through 9-months indicate improved glucose management with CGM, with increased time in range and decreased HbA1c levels.
Pediatric Moonshot
Timothy Chou, PhD
Lecturer, Stanford University Stanford, CA, USA
Our objective is to reduce healthcare inequity, lower healthcare costs and improve outcomes for children - locally and globally
Our plan is to create privacy-preserving real-time applications based on access to data in all 1,000,000 healthcare machines in all 500 children's hospitals in the world. These applications are based on a new highly de-centralized, in-the-building edge cloud service, engineered for security and privacy.
We have already demonstrated real time ultrasound image sharing across 6,000 miles and two leading pediatric hospitals. We are in the final stages of implementing the world’s first AI laboratory for children’s medicine.
The fundamental architecture includes all healthcare machines in any location. We have done very little work in endocrinology and would like to introduce this community to the potential of building privacy- preserving real-time endocrinology applications.
Personal Continuous Glucose Monitor (CGM) Accuracy During Hospitalization
Adrian G. Dumitrascu, MD; Diego De La Cruz; Michelle F. Perry, RN, CDCES; Rebecca J. Boone, APRN, DNP; Colleen T. Ball, MS; Ana-Maria Chindris, MD; Razvan M. Chirila, MD; Jessica R. Wilson, MD, MSCI
Mayo Clinic in Florida Jacksonville, Florida, USA
To assess the accuracy of patients’ personal Continuous Glucose Monitors (CGM) worn in the hospital.
We studied adult patients with diabetes mellitus (DM) that brought their personal outpatient CGM at the time of hospital admission. Patients continued to wear the CGMs, while having their blood glucose checked with point of care (POC) glucometers. CGM accuracy, clinical reliability, CGM value correlation with labs and vital signs were assessed.
We analyzed 2444 CGM-POC pairs from 74 patients (37 with Dexcom G6 and 37 with Abbott Libre 1 and 2 CGMs) across 138 hospitalizations. For all pairs, the mean absolute relative difference (MARD) was 13.4%. MARD was 22.0 % for POC glucose < 70 mg/dl. For POC values in range, grade 1, and grade 2 hyperglycemia, MARD was 14.5%, 11.3% and 9.7% respectively. G6 sensor had a MARD of 14.0% and Libre’s MARD was 11.8%.
On the Error Grid Analysis 99% of CGM values were in the “safe” zones: zone A (77.7%) and zone B (21.3%). For G6 and Libre sensors, safe zones were at 98.8% and 99.3% respectively. There was no correlation between ARD and daily mean arterial blood pressure, hemoglobin level, glomerular filtration rate and pulse oximetry (Spearman’s correlation coefficient <0.1).
If a moderate correction scale for hyperglycemia > 140 mg/dL were utilized, the difference between the mean insulin dose would be negligible (4.475 units for POC versus 4.710 units for CGM)
We found that among hospitalized patients with DM and glucose > 70 mg/dl, patients’ personal CGMs had adequate accuracy. The majority of CGM values were in the safe zones. Vital signs and laboratory values did not interfere with CGM accuracy.
Prediabetes: A Quality Improvement Project Aimed at Slowing the Disease Progression
Maryjel Espinoza, BSN, RN
Baylor Louise Herrington School of Nursing Waco, Texas, USA
The purpose of this Doctor of Nursing Practice (DNP) quality improvement (QI) project was to determine if implementing a prediabetes (PD) management protocol based on the American Diabetes Association (ADA) guidelines in a small primary care clinic is effective in preventing progression of PD in adults diagnosed with prediabetes.
The PD Lifestyle Management Protocol included ADA recommended guidelines on glucose, weight, exercise, and diet which was taught by the primary nurse practitioner. Each participant received a glucose monitor kit. Participants logged weekly fasting glucose, minutes of exercise, weight, and dietary intake for 12 weeks.
Of the 65 patients in the clinic diagnosed with PD, seven patients agreed to participate in the project—the sample group ethnicity comprised of White (71.4%) or Hispanic (28.6%). The sample was 57.1% female and 42.9% male. The mean age was 50.29; the mean weight was 196.14 lbs; the mean fasting glucose was 108.57 mg/dl, and the mean A1C was 6.0. The fasting glucose decreased by an average of 16 mg/dL (p=0.004, 95% CI; 7.59, 25.834). Average glucose reading for all participants was around 91.86 mg/dL. There was a statistically significant difference in weight average of around 9.6 pounds (p=0.009, 95% CI: 2.506, 3.440). There was an average of approximately ½ to 1 pound loss weekly. There was a mean increase in weekly exercise time of approximately 134 minutes throughout the 12 weeks. Qualitative data regarding diet showed decreased intake in carbohydrates and sugar.
Implementation of a PD management protocol in a small primary care clinic shows improvement of glucose levels, weight, exercise time, and dietary intake.
Burden of Using Technologies for Diabetes Self-Management
Sherecce A Fields, PhD; Rachel Smallman, PhD; Kianna Arthur, BS; Samantha Philip, MA; Kirsten Yehl, PhD; Erik Shoger, MBA; David Kerr MBChB, DM, FRCP, FRCPE
Texas A&M University College Station, TX, USA
This study aimed to understand current devices being used by people with diabetes and the perceived burden associated with their use to support self-management.
In collaboration with Association for Diabetes Care Educators and Specialists (ADCES), 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 (T1A; 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).
Participants with T1A were more likely to use automated means or smartphones to track their blood glucose. Nearly 25% of T2Ds not on insulin reported not tracking their blood glucose levels.
Approximately 30%, regardless of diabetes type, used paper logs to track finger stick blood glucose levels. T1A were the most likely to strongly agree that managing their diabetes would be much harder without access to diabetes apps. Two-thirds of all people with diabetes did not believe diabetes apps place extra burdens on their lives. Individuals with T2D using a Pump or MDI were more likely to wish there were additional apps to help them manage their diabetes. In addition, demographics such as age, race/ethnicity and gender significantly influenced the likelihood of being advised to use, actual use, or liking to use these different technologies.
Participants with T1D are more likely to use technological & app-based solutions to manage their diabetes compared to other groups. A significant portion of individuals still use paper-based methods for tracking their blood glucose. Technology and app developers should consider solutions more focused to those with T2D and those in certain demographic groups.
Development of a Breath-Based Glucose Estimation Algorithm
Frank Flacke, PhD; Laurent Cazals; Clarisse François-Marsal, MS; Marie-Blanche Arhainx; Sandrine Isz, PhD; Benoit Lepage, MD, PhD; Pierre Gourdy, MD, PhD
BOYDSense Toulouse, France
BOYDSense is developing a noninvasive breath based analytical platform to detect or measure disease related volatile organic compounds (VOCs). The first product will estimate glucose values from exhaled breath for people with type 2 diabetes (T2D). In a proof-of-concept study there were several VOCs identified to be correlated with glucose values. This 2-phase study, conducted at the CHU Toulouse, was designed to development and test a glucose estimation algorithm.
In this 2-phase study we enrolled 130 people with T2D, all subjected to a standardized meal challenge and predefined measurement timepoints during the 3h post-meal observation period. At each sampling timepoint, 3 breath analysis and venous reference values were collected. In phase-1 (n=100) data were collected for algorithm development. The phase-2 (n=30) was a prospective evaluation of the algorithm. All patients utilizing the breath analyzer participated also in a questionnaire.
In total 612 data pairs were collected in phase 2, with 204 pairs for each breath. For each breath, a Consensus Error Grid Analysis (C-EGA) for T2D and MARDs were calculated. The following C-EGA- distributions (zone-A/zone-A+B) were calculated: breath-1 82.8%/99.5%, breath-2 78.4%/100%, breath-3 85.3%/ 99.5%. Respective MARDs of 20.9%, 19.6% and 19.2% were estimated. 100% patients found the device not complicated to be used.
The BOYDSense breath analyzer demonstrated in a prospective evaluation of the first algorithm iteration 99.5-100% of paired readings in zones-A+B of the T2D C-EGAs and MARDs at about the same level as the first marketed CGMs. The prototype worked highly reliable, and the patients found the device not to be complicated. New studies are already designed to further enhance the performance of the device.
Comparative Effectiveness of Hybrid Closed Loop Automated Insulin Delivery Systems among Patients with Type 1 Diabetes
Sara Yi Ling Folk, BS; Kathleen Dungan, MD, MPH
The Ohio State University College of Medicine Columbus, OH, United States of America
The objective is to compare glycemic outcomes among patients with type 1 diabetes (T1D) initiating hybrid closed loop (HCL) insulin pumps in a real-world setting.
This was a retrospective study of patients from an academic Endocrinology practice between 01/01/2018- 11/18/2022. Inclusion criteria were diagnosis code for T1D, >18 years of age, new to any HCL system (Medtronic 670G/770G [MT], Tandem Control IQ [CIQ], or Omnipod 5 [OP5]) and availability of a pump download within 3 months. Primary outcomes included % time in range 70-180 mg/dl (TIR) at 90 days and HbA1c at 90-180 days.
Of the 176 participants, there were 47 MT, 74 CIQ, and 55 OP5. Median age was 41 (interquartile range [IQR] 30, 54) years, 39% male, 88% white, diabetes duration 22 (IQR 14-34) years. Baseline characteristics were similar, except more patients who started MT were on CGM and an insulin pump. Change in HbA1c was -0.1 (-0.8,0.3), -0.6 (-1.1, -0.15), and -0.55 (-0.98, 0) for MT, CIQ, and OP5 respectively (p= 0.04). However, within group change in A1c was statistically significant in all three groups. The %time in auto mode was 80 (54, 89), 95 (89, 98), and 95 (82, 100)% respectively (p<0.0001).
TIR was 70 (57, 76), 67 (59, 75), and 68 (60, 76)% (p=0.95) at 90 days while %time < 70 mg/dl was 2 (1, 3), 1 (0,2), and 1 (0, 1)% respectively (p=0.002).
There were significant reductions in HbA1c in all HCL systems favoring CIQ and OP5. TIR was no different between groups but %time below 70 mg/dl was highest with MT. A1c reductions as well as %TIR were related to %time in auto mode.
Type 2 Diabetes Diagnosis and Bacterial Marker Identification Through Gut Microbiota
Prisha Goyal
Khan World School Bhopal, Madhya Pradesh, India
Type 2 diabetes mellitus is a chronic metabolic condition that diminishes glucose uptake in the body due to insulin resistance. Connections have been proposed between Type 2 diabetes and the human gut microbiome; as there are 422 million people with Type 2 diabetes globally, insights derived in this area will benefit a large number of people. The objective of this study was to investigate the emerging area of gut biomarkers by employing machine learning, another technology poised to make an impact in medical research.
Analysis of the associations between Type 2 diabetes and the human gut microbiome was carried out by obtaining a dataset of the microbiotic profiles (n=446), with an equal number of people with diabetes (n=223) and people without diabetes (n=223). The data was divided into training and testing sets in a 50-50 ratio and a binary XGBoost classifier was trained to distinguish between the diabetic and non-diabetic classes on the basis of bacterial expression levels.
An accuracy of 92.40%, sensitivity of 94.59%, and specificity of 88.33% was obtained. The top five bacteria for predicting Type 2 diabetes were Clostridium symbiosum, Coprococcus comes, Streptococcus anginosus, Clostridium nexile and Veillonella atypica. Three of these are novel results, based on statistical significance testing through the Wilcoxon rank sum test.
This research could help utilize machine learning for quicker, more accurate, noninvasive diagnosis of Type 2 diabetes. Other researchers could validate the bacterial associations found in this study and if possible, identify the causative mechanisms of action. It could also help discover newer preventive and therapeutic medicines for Type 2 diabetes and design probiotics for personalized treatment of each patient based upon their bacterial microbiome profile.
Advances in Automated Insulin Delivery with The Medtronic 780G System: The Australian Experience
Bella Halim, MBBS; Mary B. Abraham, MBBS, DCH, MD Paediatrics, FRACP, PhD; Georgina Manos MBBS; Sara Vogrin MBBS; Richard MacIsaac BSc (Hons), PhD, MBBS, FRACP; Elif I. Ekinci MBBS, FRACP, PhD; Alicia Jenkins MBBS, MD, FRACP, FRCP; John Shin PhD, MBA; Robert A. Vigersky MD; Tim Jones MBBS, DCH, FRACP, MD; David N. O’Neal MD, FRACP, PhD
St Vincent’s Hospital Melbourne, University of Melbourne Fitzroy, VIC, Australia
Automated Insulin Delivery (AID) devices have evolved, aiming to improve glucose control and user- experience. Enhancements to the MiniMedTM Medtronic 780G AID system include improvements in the AID algorithm, choice of glucose targets, automated correction bolus up to every 5-minutes, and improved user interface. We aim to compare real world data of MiniMedTM Medtronic 780G.
We conducted a retrospective analysis of CarelinkTM real world data from Australian children and adults collected between January 2020 and December 2022, comparing glycaemia and device performance of MiniMedTM Medtronic 670G/770G and 780G. All used GuardianTM Sensor 3 and had at least 10 days of sensor glucose (SG) data following AID (SmartGuardTM) activation.
There were 5676 670G/770G users and 3566 780G users. A greater proportion of 780G users met recommended CGM targets across all age groups, which included glucose management indicator (GMI) < 7% (38.3% vs. 59.7%; p<0.001), time in range (TIR) >70% (43.4% vs. 64.1%; p<0.001) and time below range (TBR) <4% (88.9% vs. 90.5%; p=0.016).
Compared with 670G/770G, a greater time was spent in SmartGuardTM using 780G with fewer exits (3.8±2.3/week vs.1.4±1.3/ week; p<0.001), and fewer capillary glucose readings (SMBG) were recorded (6.2±2.5/day vs. 3.8±1.6/ day; p<0.001). Total daily insulin dose did not differ (51.5±25.6 U/day vs.
52.8±28.0 U/day; p=0.06). Among 1051 670G/770G users who transitioned to 780G, there was improvement in TIR (70.0±10.7% vs. 74.0±10.2%; p<0.001) without significant change in TBR (2.0±1.6% vs. 2.0±1.7%; p=0.7704).
Compared with the first-in-market Medtronic 670G/770G AID systems, a significantly greater proportion of 780G users meet GMI and TIR targets while reducing hypoglycaemia and SMBG readings. When switching from 670G/770G to 780G, there was further glycemic benefit.
Improvements in the Accuracy of a Pre- Commercial Non-Invasive Wearable Glucose Monitor Through the Use of AI
Consuelo Handy, BSc; Hassan Kayali, BSc, MSc; Prof. Stephen Charles Bain, MD; Prof. Steve Luzio, PhD; Muhammad Rafaqat Ali Qureshi, PhD; Bradley Love, PhD; Nuno Miguel Machado Costa da Silva, BSc; Luis Filipe Rocha Maia Ferreira, BSc; Kathie Wareham, MPH; Lucy Barlow; Gareth J. Dunseath, PhD; Joel Crane, PhD; Isamar Carrillo Masso, PhD; Julia A. M. Ryan, BA; Mohamed Sabih Chaudhry, PhD
Afon Technology Ltd, Unit 670, Castlegate Business Park, Caldicot Road, Caldicot, Monmouthshire, NP26 5AD, United Kingdom
A noninvasive, wearable, continuous glucose monitor would be a major advancement in diabetes therapy. This trial investigated a novel noninvasive glucose monitor which analyzes spectral variations in electromagnetic waves reflected from the wrist.
A single-arm, open-label, experimental study compared glucose values from a prototype investigational device with laboratory glucose measurements from venous blood samples (YSI) at varying levels of glycemia. The study included 20 male and female participants with type 1 and type 2 diabetes (age range = 34-76 years).
The study comprised of three five-patient cohorts testing an improved prototype to estimate the accuracy of our technology by comparing the predicted glucose levels vs the patient-specific actual blood glucose levels over time, as measured using a YSI and a commercially available CGM as comparators, including before and after glucose challenge. The co-primary endpoints in all trial stages were median and mean absolute relative difference (MARD) calculated across all data points.
Each cohort was used to train a neural network, using a leave-one trial-out approach to predict the left-out trial. This was done for each trial using all the data. Adding each successive cohort increased the accuracy of the algorithm. Using just Cohort 1, we achieved an overall MARD of 21%, using cohorts 1 and 2 an overall MARD of 15%, using cohorts 1, 2 and 3 and overall MARD of 13%. This shows that not only will the device predict glucose, but as the neural network is given more data, the accuracy of the predictions is improving.
This study concluded that the novel noninvasive continuous glucose monitor tested was capable of detecting glucose levels.
NonLinear Association between Blood Glucose Levels and Walking in an Integrated Digital Health Platform for Diabetes Management
Yifat Fundoiano-Hershcovitz PhD; Inbar Breuer Asher MPh; Halit Kantor, BSc; Sandy Rahmon BA; Ephraim Bachar BSc; Omar Manejwala MD; Pavel Goldstein, PhD
Dario Health Caesarea, Israel
The effect of physical activity on blood glucose levels is widely known to be a cornerstone in the management of type 2 diabetes. Digital therapeutics platforms for chronic condition management are aiming to generate behavioral changes to improve patient’s clinical outcomes. The purpose of this study is to investigate the association between monthly aggregated blood glucose measurements and walking activity (number of steps) measured on a single digital therapeutic platform.
In this retrospective real-world study, a cohort of 987 platform users with Type 2 diabetes and pre- diabetes, who regularly tracked their blood glucose levels for 12 months using the Dario digital platform was evaluated. The association between blood glucose levels and the number of steps was examined over time. A piecewise linear mixed effects model was applied to test the trajectories over time of monthly average blood glucose and monthly average number of steps a day in two time periods defined by previous research as well as to test nonlinear association between them.
Analysis revealed that during the first 4 months, there is a positive trend of monthly average steps (B=0.03, P<.001) while there is a negative trend of blood glucose levels (B= -0.01, P<.001) in the same time frame. Significant improvement in monthly average blood glucose was observed in users with at least 400 steps a day (B= -0.02, P<.001), while for those with less than 400 steps a day, there was no significant change.
This study sheds light on the importance of physical activity in diabetes management. Our findings highlight the potential in promoting physical activity using a diabetes-tracking app to improve clinical outcomes.
Intravenous Continuous Glucose Monitoring in Hospitalized Patients in a Critical Care Unit: Safety and Efficacy of a Novel Intravascular Continuous Glucose Monitoring System - A Pilot Trial
Daniel Hochfellner, MD; Tina Poettler; Michael Schoerghuber, MD; Edita Lukic, PhD; Ameli Yates, MD; Ingeborg Keeling, MD; Daniel Zimpfer, MD, MBA; Laura Roubik, BSc; Hesham Elsayed, MD; Felix Aberer, MD, PhD; Francesca Berti, PhD; Fausto Lucarelli, PhD; Francesco Valgimigli, PhD; Julia Mader, MD
In critically ill patients with and without diabetes mellitus, glucose swings represent a significant risk. Whilst patients on general wards might benefit from continuous glucose monitoring (CGM), it is withheld from critically ill patients as accuracy data is lacking. This leads to frequent blood glucose measurements increasing workload for nursing staff. Staff often fear hypoglycemia more than consequences of hyperglycaemia, resulting in insufficient glycemic control and associated complications.
In this single-center, open-label study, we evaluated the accuracy and safety of an intravascular CGM system (GlucoMen®Day ivCGM system, A. Menarini Diagnostics, Florence, Italy) in patients with and without diabetes hospitalized in a cardiothoracic surgery intensive care unit. Intravascular microdialysis- based CGM was performed by accessing patient blood stream through a standard iv-line, inserted postoperatively in one of the patients' forearms. The investigation period was up to 72 hours postoperatively. Data derived from ivCGM were paired with glucose data measured from arterial or venous samples using a custom blood gas analyzer. Diabetes management was performed by routine staff without interference by study procedures.
Of the 35 patients included in the study (age 65±10 years, female 25.7%, BMI 28.2±5.4 kg/m², diabetes mellitus: 28.6%), 29 completed the study and 28 were evaluable for final analysis. Compared to standard glucose measurements, during a median run time of 60.2 (IQR 42.0-67.3) hours, the system achieved a mean absolute relative difference (MARD) of 8.7±7.8%. In the Clarke Error Grid (CEG), 100% of the data pairs were in the clinically acceptable zones A and B (90.3% of data in A). No device related adverse events were observed.
IvCGM provided good accuracy over a wide glycemic range without any device related adverse events. In future trials, the usefulness of ivCGM to steer glucose management will be tested.
Structured CGM Education Delivered via Mobile Phone Is Associated with Increased Time in Range: Interim Analysis
Melissa P. Holloway, MSt; Thomas P. Grace, MD; Vassilis Karamanis, MSc; Christopher G. Parkin, MS
SmartStart Health Ltd London, UK
SmartStart CGM is a structured educational program for current users of continuous glucose monitoring (CGM) designed for use on a mobile phone. In this proof of concept study we investigated the effects of SmartStart CGM on acquired knowledge and established glycemic metrics in a cohort of adults with diabetes on intensive insulin therapy. Planned enrollment for this study is 40 adult participants. We report findings from the first 17 participants to complete the program.
SmartStart CGM presents basic and advanced information about using CGM and interpreting CGM data. After enrollment, participants answer 20 “welcome check” questions to assess baseline knowledge. Upon completion of 7 interactive content modules, they answer the same 20 questions as a “final check.” Participants’ time in range 70-180 mg/dL (%TIR) and time below range <70 mg/dL (%TBR) were obtained at baseline and 6 weeks post- completion. Changes in CGM knowledge and glycemic metrics from baseline were assessed.
Average duration of CGM use prior to enrollment was 35±17 months. The average CGM knowledge score rose from 39% on the “welcome check” to 87% on the “final check.” At 6 weeks post-completion,
%TIR increased in all participants with %TIR of <70% at baseline (n=8); 62.5% showed a clinically meaningful ≥5% increase. Among participants with baseline %TIR of ≥70% (n=9), 77.78% maintained/increased %TIR by up to 4%, and 22.22% showed an increase of ≥5%. All participants (100%) achieved the %TBR goal of <4%. The percentage of participants reaching ≥70% TIR and <4% TBR increased from 29% to 65%.
This interim analysis strongly suggests that completion of SmartStart CGM was associated with increased CGM knowledge, increased %TIR and reduced %TBR.
Development of Novel Ketone Biosensing Receptors for Electrochemical Continuous Monitoring
Bryant J. Kane, BS; Mika Hatada, PhD; Junko Okuda-Shimazaki, PhD; Joseph Kerrigan Jr., BS; Koji Sode, PhD
Joint Department of Biomedical Engineering, The University of North Carolina at Chapel Hill and North Carolina State University, Chapel Hill, NC, USA
β-hydroxybutyrate (BHB) is an important biomarker for people with insulin deficiency, as they can unknowingly develop diabetic ketoacidosis, a life-threatening complication. Current electrochemical measurement systems for BHB are limited by the nature of BHB dehydrogenase enzymes. The activity of these enzymes against BHB is dependent on the availability of NAD+/NADH cofactor, which makes them difficult to configure for continuous operation. Solute binding proteins (SBPs) are receptors which bind their targets reversibly in the absence of cofactor without initiating catalysis and have been used for construction of effective continuous biosensing systems. Before the work described herein, a similar system for electrochemical continuous ketone monitoring (CKM) was not feasible from lack of an existing SBP.
We selected candidate BHB-binding SBPs through analysis of sequence and structural homology, docking, and molecular dynamics simulations. The top candidates were expressed recombinantly in Escherichia coli and purified from cell lysates. We checked binding activity against BHB and acetoacetate using fluorescence spectroscopy.
Across a titration regime in the nanomolar-micromolar range, we observed concentration-dependent change of intrinsic tryptophan/tyrosine fluorescence and concentration-dependent change of current during pulsed amperometry experiments. Appreciable signal change was not observed from addition of structurally similar interferents, such as amino acids, indicating high SBP specificity.
In this study we report novel SBPs for BHB, and their application for CKM. To the best of our knowledge, this is the first report which suggests the existence of soluble binding proteins which are specific for ketone bodies. Our discovery has facilitated the development of an innovative, NAD/NAD+ dependent, CKM platform. We envision that with future characterization, a sensor harboring this SBP can comprise a solution for CKM which circumvents the challenge of using NAD+/NADH dependent enzymes.
Two Center Real-World Evaluation of Spotlight-AQ Improves Glycaemic Control and Reduces Diabetes-Related Distress
Ryan Kelly, BSc, MBA; Richard I. G. Holt, MD, PhD; Naresh Kanumilli, MD; Nicola Milne, RN; Matthew Owen, MD; Catherine Whitrow, RN; Ethan Barnard, BSc, John Welsh, MD, PhD; Katharine Barnard-Kelly, Ph
Spotlight-AQ Ltd., Fareham, Hampshire, UK
To determine the impact of specific tailored, patient-centred onboarding of rtCGM on physical and mental health outcomes for adults with type 2 diabetes (T1D). The Spotlight-AQ platform consists of biopsychosocial pre-clinic assessment, immediate graphic results presentation and mapped care pathways for unmet needs.
Two-center real-world evaluation of Spotlight-AQ platform in primary care practices with adults with T2D. Participants completed Spotlight-AQ tool at baseline and wore Dexcom One rtCGM systems for three months. Primary outcome: HbA1c change; secondary outcomes: diabetes distress score (DDS) and glucose monitoring satisfaction.
Twenty-three participants were recruited across two primary care centres in Manchester, UK. Of the 21 who completed, eight were female, mean age was 64 (range, 48-86) years, 20 were white, and all used injectable therapy. The mean HbA1c fell by 14.95 mmol/mol (range, 4-71 mmol/mol) (p<0.001). Baseline diabetes-related distress was moderate, reducing significantly at follow-up for total score and all subscales. Women reported greater total distress, emotional burden, physician related burden and regimen-related distress at baseline than men (p=ns). Men reported significantly greater interpersonal distress at baseline. At follow-up, no significant differences between total distress score, emotional distress, or regimen distress were observed across genders. Women reported lower physician burden and interpersonal distress at follow-up than men (both p<0.001). Women reported significantly greater improvements across total DDS score, emotional & physician burden and regimen distress (all p<0.0001), whereas men reported significantly greater improvement in interpersonal distress (p0.001). GMSS scores did not improve overall or within worthwhileness or openness subscales. Emotional and behavioral burden fell significantly (p<0.0001).
Spotlight-AQ was associated with significant HbA1c reductions and mental health improvements for adults with T2D. These benefits conferred a cumulative improvement with use of Dexcom One rtCGM.
Using Early Engagement Data from a Digital Health Solution to Predict Future Glycemia Risk Index (GRI)
Abhimanyu Kumbara, MS; Junjie Luo, MS; Anand Iyer, PhD; Mansur Shomali, MD, CM; Guodong “Gordon” Gao, PhD, MBA
Welldoc Columbia, MD, United States
We have previously demonstrated that combining a digital health solution and CGM devices supports improvements in glucose management. Dense data from CGM devices allows the calculation of a stable and composite metric like glucose risk indicator (GRI), which can be important in predicting future health outcomes. We studied the use of early digital engagement data to predict future GRI.
A real-world data set of 499 CGM users with type 1 and type 2 diabetes (T1D and T2D) was created. Baseline was defined as the first 30 days of use from registration. The prediction period was between days 70 and 90 from baseline. Users with >70% sensor wear time in the prediction period were included in the prediction dataset (n=304). The GRI prediction variable was categorized as good if the GRI score was <= 40 and bad if the GRI score was >40. A Gradient Boosting Classifier (GBC) was used to predict future GRI outcomes in three population subgroups: Overall (n=304), T1D only (n=125), and T2D only (n=140).
The GBC was highly accurate in predicting binary, future GRI outcome in all subgroups. The overall model accuracy was 0.83, and 0.88 and 0.80 for the T1D and T2D subgroups respectively. All three models had AUC scores >0.9.
These data demonstrate the potential for early engagement data from a digital health solution to predict future GRI outcomes. Predicting GRI may help health plans and care teams to design highly-personalized treatment plans to optimize glucose management at both individual and population levels. This work is also foundational to leveraging real-time data like CGM to evolve digital health AI capabilities.
Developing an Artificial Intelligence Based Algorithm for Predicting Glucose for Patients with Type 1 Diabetes through Multiple Continuous Life-Log Variables
Joonyub Lee MD, PhD; Moon Tae Hwang; Jin Yu MD; Kun-Ho Yoon MD, PhD
Division of Endocrinology, Department of Internal Medicine, Seoul St. Mary's Hospital, College of Medicine, The Catholic University of Korea, Seoul, Korea
This study aimed to develop a machine learning-based algorithm capable of predicting glucose levels in type 1 diabetes patients using multiple continuous life-log variables.
In this ongoing prospective observational study, eighteen type 1 diabetes patients from two diabetes centers (Seoul St. Mary’s Hospital and Yeungnam University Hospital) in South Korea were enrolled. Each patient was provided with EOPatch (EOflow) insulin pumps, Free Style Libre 2 (Abott) continuous glucose monitors, Dofit pro band (Mediplus Solution) activity tracker, and a food tag AI (KT). Along with the EMR data, four types of continuous life-log variables were obtained through the Korea Health Partners' patient-physician communication app. The collected data were analyzed using a recurrent neural network based Long Short-Term Memory algorithm.
The Mean Absolute Error (MAE) for 30 minute glucose prediction decreased from 27.55 mg/dL to 15.63 mg/dL when dietary, activity, and insulin information was incorporated. The MAE was 7.12 mg/dL at 5 minutes, 15.63 mg/dL at 30 minutes, and 35.12 mg/dL at 2 hours. Among the continuous life-log variables (diet, insulin, and physical activity), physical activity emerged as the most influential variable affecting blood glucose levels at the 30-minute mark. An increase in calorie consumption led to a step- wise decrease in blood glucose at 30 minutes (20 Kcal: -6.1mg/dL, 45 Kcal: -15.04 mg/dL, 90 Kcal: - 34.35 mg/dL, 180 Kcal: -80.25 mg/dL).
The machine learning algorithm based on multiple continuous life-log data significantly improved the accuracy of glucose prediction. This data is poised to be a valuable asset in the development of digital twin technology for patients with type 1 diabetes.
AI-Powered Clinical Trials: Emulating Real- World GLP-1 Efficacy with Synthetic Patient Populations using Causal Effect Learning
Calum R. MacLellan, MEng; Conor McKeag, MBChB, MD; Hristo Petkov, BSc; Feng Dong, PhD; David Lowe, MBChB,MD; Roma Maguire, PhD; Sotiris Moschoyiannis, PhD; Jo Armes, PhD; Simon S. Skene, PhD; Dr. Christopher Sainsbury, MBChB, MD
Department of Computer and Information Sciences, University of Strathclyde Glasgow, Lanarkshire, United Kingdom
Randomized controlled trials (RCT) remain the gold standard for quantifying treatment effects but are expensive to conduct and often limited in scope by design. We introduce a new paradigm to perform virtual clinical trials using a generative artificial intelligence (AI) algorithm that infers effect size, which can potentially extrapolate the results of RCTs to wider populations through simulation. We focus on therapies for glycemic control in type 2 diabetes (T2DM) and validate the virtual trials by comparing the predicted impact on HbA1c of glucagon-like peptide-1 agonists (GLP-1) with its true efficacy as established in the LEAD-5 clinical trial.
Observational training data comprised pre- and post-treatment outcomes for 5,476 people with T2DM. We considered three treatment arms to reflect LEAD-5: GLP-1, basal insulin, and placebo. After training ouralgorithm, we performed virtual trials (N=60 per arm) by sampling 232 synthetic patients for each arm according to the LEAD-5 inclusion criteria and predicting post-treatment outcomes. We used difference- in-differences (DiD) for pairwise comparisons between each arm: our goal was to match LEAD-5 by demonstrating a significant DiD in post-treatment HbA1c reduction for GLP-1 compared to both basal insulin and placebo.
We found a significant difference in HbA1c reduction for GLP-1 vs basal insulin (DiD: -1.21; 95%CI: - 1.42, -1.00; p<0.001) and GLP-1 vs placebo (DiD: -2.58; 95%CI: -2.78, -2.37; p<0.001) in our synthetic populations, which agrees with LEAD-5 (Liraglutide vs glargine: -0.24% difference; 95%CI: 0.08, 0.39; p=0.0015, and Liraglutide vs placebo: -1.09% difference; 95% CI: 0.90, 1.28; p<0.0001).
AI-powered clinical trials can match real-world RCTs in important clinical measurements. Our algorithm is specialty agnosticand can explore counterfactual questions, making it broadly suitable as an assistive tool for precision medicine and clinical decision-making.
Statistical Software Packages and Algorithms for Analysis of Continuous Glucose Monitoring Data – A Systematic Review
Mikkel Thor Olsen, MD; Carina Kirstine Klarskov, MD, PhD; Katrine Bagge Hansen, MD, PhD; Ulrik Pedersen-Bjergaard, MD, PhD; Peter Lommer Kristensen, MD, PhD
Department of Endocrinology and Nephrology, Copenhagen University Hospital – North Zealand Hilleroed, North Zealand, Denmark
Continuous glucose monitoring (CGM) provides glucose level measurements every 1-15 minutes and is widely used in clinical and research contexts. Handling large amounts of CGM data is challenging.
Statistical packages and algorithms might ease the time-consuming and error-prone process of the manual calculation of CGM metrics and contribute to the standardization of CGM metrics defined by international consensus. The objective of this systematic review is twofold: (1) to summarize statistical packages for CGM data analysis and (2) to summarize statistical algorithms not available in these statistical packages.
A systematic literature search in PubMed and EMBASE was conducted on November 16th, 2022, in addition to a free-text literature search and reference checking in the included literature up until April 21st, 2023. Articles in English and Danish were included. This systematic review is registered with PROSPERO (CRD42022378163).
A total of 3687 references were screened and 44 references were included. We identified 21statistical packages for the analysis of CGM data. The statistical packages could calculate a great amount of 2022 internationally consensus CGM metrics in addition to other historical CGM metrics. Further, 20/21 (95%) of these packages were freely available. We also identified 23 statistical algorithms. The algorithms could be divided into three subgroups depending on their intended use: (1) CGM data reduction (e.g., clustering of CGM data), (2) composite CGM outcomes, and (3) other CGM metrics.
This systematic review provides detailed tabular and textual up-to-date descriptions of the contents of statistical packages and statistical algorithms for retrospective analysis of CGM data.
A New Portable HbA1c Meter—Red Blood Cell Deformability–Based HbA1c Test
Eunyoung Park, PhD; Seungjin Kang, ME; Ung Hyun, PhD
Orange Biomed Seoul, Korea
In the United States, many people still do not benefit from the HbA1c test due to its limited accessibility. Alternatively, a portable HbA1c meter, which lay users can easily operate, could be a possible solution to monitor their HbA1c levels much conveniently. However, the current HbA1c test requires precise sample volume, demands complicated operation and maintenance, and can be distorted the test results by hemoglobin variants, especially for people of color. Therefore, we suggest an innovative method for the A1c test that solves these problems and makes the A1c test accessible to everyone, anywhere.
Physiologically the RBCs deformability is inversely correlated with the amount of glycated hemoglobin. Taking advantage of this relationship, we have developed an innovative HbA1c meter that estimates HbA1c by measuring the deformability of individual RBCs. This novel meter utilizes a microfluidic channel with dimensions narrower than those of RBCs, enabling the evaluation of RBC deformability by detecting the velocity of individual RBCs as they pass through the fluidic channel.
We successfully detected thousands of RBCs velocity within minutes and verified that the RBC deformability was inversely correlated with HbA1c level. And from the results, we successfully converted the HbA1c level from estimated RBC deformability data.
We successfully develop new HbA1c meter that is free from sampling error and maintenance. And this novel technology also free from interference due to hemoglobin variants, providing accurate results despite one’s racial/ethnic background.
DR-CIB: an Algorithm for the Preventive Administration of Corrective Insulin Boluses in Type 1 Diabetes based on Dynamic Risk Concept and Patient-Specific Timing
Elisa Pellizzari, MSc; Francesco Prendin, PhD; Giacomo Cappon, PhD; Giovanni Sparacino, PhD; Andrea Facchinetti, PhD
University of Padova Padova, Italy
Several strategies to reduce the duration of post-prandial hyperglycemia in type 1 diabetes (T1D) open- loop therapy have been developed in the recent years. Although these heuristics proved to be valid options accounting for continuous glucose monitoring (CGM) trend, they present some limitations and lack of personalization, calling for a more efficient solution.
We developed DR-CIB, a novel algorithm for post-prandial corrective insulin bolus (CIB) suggestion based on a preventive trigger threshold exploiting the “risk of hyperglycemia” and a personalized CIB timing retrieved from patients’ specific glucose-insulin dynamics. DR-CIB has been assessed on a dataset consisting of 49 daily CGM traces recorded in real-life conditions using ReplayBG, a novel digital twinning tool that allows a retrospective assessment of alternative insulin therapies using real data. As comparators we evaluated state-of-the-art approaches proposed by Aleppo (AL), Bruttomesso (BR), and Ziegler (ZI). Efficacy of glucose control was quantified by temporal, risk, and hyperglycemic event metrics.
Compared to literature methods, DR-CIB significantly reduces time spent in hyperglycemia when compared to AL and BR (33.52% vs 39.76% and 36.32%, respectively); significantly reduces daily injected insulin (5.97U vs 7.5U), glycemia risk index (37.78 vs 40.78) and time spent in hypoglycemia (75th percentile from 10.23% to 1.74%) when compared to ZI, resulting overall in a safer solution.
We proposed DR-CIB, a dynamic risk-based algorithm which allow preventive actions for ahead-in-time management of hyperglycemic events and overcome some literature limitations proposing a patient- specific timing for CIB. DR-CIB proved to be a valid alternative to the most recent heuristic literature guidelines reducing the time spent in hyperglycemia and the hyperglycemic events duration without increasing the time below hypoglycemic threshold.
Continuous Glucose Monitoring with an Osmotic-Pressure Based Continuous Glucose Sensor: Results of the First Human Pilot Study
Andreas Pfützner, Mina Hanna, Nicole Thomé, Boris Stamm, Joacim Holter
Pfützner Science & Health Institute Mainz, Germany
The osmotic-pressure-based Sencell glucose sensor technology (Lifecare AS, Bergen, Norway) is expected to provide a close to linear correlation between the raw sensor signal and the glucose concentration and a very long duration of use (of up to 6-12 months or longer). The final device is planned to have the size of a grain of rice and to be implanted employing wireless energy and data transfer.
In a first clinical proof of concept study in humans, a wired version of the core sensing technology was employed, which was embedded into a 4 mm needle and inserted into the abdominal subcutaneous tissue. The study was conducted to collect first human proof-of-concept performance data for algorithm development during meal experiments and for further device optimization. The raw data was analyzed after one-point calibration and minor trend correction in comparison to the Statstrip blood glucose meter and commercially available CGM glucose sensors.
Nine participants (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 with observation periods between 2 h and up to 72 h. The osmotic-pressure sensor followed glucose changes similar to the FreeStyle Libre 2 or Dexcom G7 device and 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, and 78% reached the acceptance criteria for blood glucose meters in the Bland-Altman-Analysis.
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. These results support the further development of Sencell towards a clinically usable medical device.
Dynamic Interference Test Results with Two Glucoseoxidase-Based CGM Needle Sensors
Andreas Pfützner, Hendrick Jensch, Geetham Srikanthamoorthy, Christian Kuhl, Lukas Weingärtner, Mike Grady, Steven Setford, Elizabeth Holt, Nicole Thomé
Pfützner Science & Health Institute Mainz, Germany
Little is known about the reaction of glucoseoxidase (GOD)-based continuous glucose monitoring sensor technologies to potentially interfering nutritional, endogeneous or pharmaceutical substances. Here we report on results obtained with two GOD-based commercially available needle sensors with our in-vitro dynamic intererence testing method (Pfützner et al., J. Diabetes Sci. Technol., 2022, May 13:19322968221095573).
We used HPLC pump-controlled substance gradients to expose Dexcom G6 (G6) and Freestyle Libre 2 (L2) needle sensors in a 3D-printed test cartridge to varying concentrations of potentially interfering substances at a fixed glucose concentration of 200 mg/dL. We tested 68 substances in triplicate using YSI Stat 2300 Plus as the glucose reference method. Interference was assumed if a CGM needle sensor showed more than ±10% difference from baseline with a tested substance at any given concentration. The bias observed at the maximal interferent concentration was expressed as %BOB (bias over baseline).
Interference with both sensors was seen with the following substances (BOB: G6/L2): mannose (20%/+130%), hydroxyurea (>100%/84%), %), galactose (17%/>100%), N-acetyl-cysteine (18%/11%), and dithiothreitol (-11%/46%). G6 but not L2 showed interference by acetaminophen (>100%), uric acid (33%), gentisic acid (18%), ethyl alcohol (12%), L-dopa (11%), and L-cysteine (-25%), while no influence on G6 but on L2 was observed with xylose (>100%), ascorbic acid (48%), methyldopa (16%), ibuprofen (14%), red wine (12%), and icodextrin (+10%). No interference was seen with other tested substances. In addition, the G6 sensors subsequently ceased to operate when exposed to dithiothreitol, L- cysteine, gentisic acid, and mesalazine (suspected sensor fouling, a phenomenon not observed with L2).
Employing our standardized dynamic interference test protocol, several nutritional and pharmacological substances were identified to influence the signals of G6 and/or L2. While both CGM sensors are based on a glucoseoxidase electrode technology, the interference pattern was observed to have some common molecules but was otherwise substantially different. The clinical implications of our in-vitro findings, however, must be confirmed in both cases with appropriately designed clinical studies.
Phenotyping of Type 2 Diabetes by Means of Functional Biomarkers for the Underlying Deteriorations Provides a Valuable Tool for Personalized Treatment Selection – A Discussion Paper
Andreas Pfützner, Mina Hanna, Daiva Kalasauske, Daniela Sachsenheimer
Pfützner Science & Health Institute Mainz, Germany
Type 2 diabetes (T2DM) is a highly complex disease consisting of multiple partially inherited underlying root disorders (ß-cell dysfunction, chronic systemic inflammation, vascular and metabolic insulin resistance, etc.) and often leading to multiple consecutive microvascular and macrovascular sequelae. The gluco-centric (symptomatic) focus of current treatment guidelines (HbA1c) turns T2DM into a chronic progressive disease with high probability of premature death due to macrovascular events.
Almost 15 years ago, we had identified biomarkers, which next to the well-established routine parameters (HbA1c, glucose, lipids, BMI, blood pressure) are helpful to describe the degree and severity of the underlying root deteriorations (Pfützner et al., Clin.Lab., 2008), namely intact proinsulin (ß-cell dysfunction), hsCRP (chronic systemic inflammation), and total adiponectin (insulin resistance and hormonal activity of the visceral lipid tissue). In the same year, we started to treat patients with personalized anti-diabetic drug combinations directed towards the identified phenotype (SOC+= personalized standard of care).
Until today, several hundred patients have been treated according to our SOC+ concept. In many cases, we were e.g. able to induce a long drug-free remission phase (3 months to >5 years) even in advanced patients by a 3-month temporary treatment of all underlying deteriorations in parallel (with the focus on the leading deterioration; so called “De-Escalation Treatment”, DET). When diabetes progressed again - usually indicated by an increase in intact proinsulin - we were often able to repeat DET with similar success. Our SOC+-treated patients appear to have better glycemic control, fewer secondary complications, and only very few of them have passed away so far.
We report our SOC+ concept and findings at this stage to invite our peers to try it out and make their own experience (hopefully as positive as ours). We are lacking the interest and support of the pharmaceutical industry, the healthcare systems, and of the established academic community to support and fund appropriately designed long-term randomized prospective clinical studies (RCTs) to proof the advantages of our SOC+ concept in comparison to the gluco-centric and escalating treatment guidelines. In addition, we would most likely not live long enough to personally experience the analysis and results of such studies.
However, we would like to remind all of us that evidence-based medicine in its original meaning is not only comprised of RCTs, but also on pathophysiology, physician experience, and patient preference.
Evaluating the Effect of EndoTool Utilization for Glycemic Control in Critically Ill Patients
Kinza Salim, DO; Josephine Gomes, DO; Evelyn Calderon Martinez, MD; Danya Abedeen, DO; Zachary Scheid, DO; Rabiah Riaz, MD; Lauren Ortiz; Anderson Schrader; Amy Helmuth, DNP, FACHE; Soni Srivastav, MD; Paul Chidester, MD, FACP; Anas Atrash, MD, FACP
UPMC Central Pennsylvania Harrisburg, PA, United States
To analyze patient outcomes for a comparable three-month period pre- and post-implementation of EndoToollV (ETIV) in two critical care units at UPMC Central PA
A retrospective study was performed to compare patients who were on an insulin dripfor hyperglycemiain the ICU pre-ETIV (baseline group that included 125 patients) and post-ETIV implementation (162 patients) at the UPMC Central PAfrom May to June 2022and September to December 2022,respectively. The primary outcome was the time (days) on an insulin drip. Secondary outcomes included hypoglycemic events, ICU length, length of hospital stay, and cost.
Post-ETIV implementation showed statistically significant reduction in time of insulin infusion (58.2 hours vs 41.4 hours, p=0.0008), average time to reach goal blood glucose of 140 (18.15 hours vs 4.7 hours, p < 0.0001), rate of hypoglycemic events with blood glucose < 70 (2.11% vs 0.53%, p < 0.0001%), Implementation of ETIV also demonstrated a reduction in ICU length of stay (5.3 days vs 4.76 days), hospital length of stay (11 days vs 8.6 days), cost of ICU charges per patient($42,000 vs $32,000).
Utilizing ETIV led to a lower time on insulin drip, goal blood glucose was reached in one fourth of the time as that required by the baseline group. The use of ETIV achieved a fourfold reduction in hypoglycemia (BG< 70mg/dl) and the elimination of severe hypoglycemia (BG <40 mg/dl). The efficacy of EndoTool IV on quality measures can be seen with its elimination of severe hypoglycemia in management of hyperglycemia in the ICU. The use of a computerized dosing algorithm such as ETIV can help improve patient safety and clinical outcomes.
Utilization of a Computerized Dosing Algorithm toImprove Management in Diabetic Ketoacidosis
Zachary Scheid, DO; Kinza Salim, DO; Josephine Gomes, DO; Danya Abedeen, DO; Rabiah Riaz, MD; Evelyn Calderon Martinez, MD; Lauren Ortiz; Anderson Schrader; Amy Helmuth, DNP, FACHE; Soni Srivastav, MD; Paul Chidester, MD, FACP; Anas Atrash, MD, FACP
UPMC Central Pennsylvania Harrisburg, PA, United States
The study's primary objective was to assess the impact of a computerized dosing algorithm, EndoTool IV (ETIV), on the cost and clinical outcomes of managing Diabetic Ketoacidosis (DKA), in comparison to the existing computer-based order set.
ETIV, providing dosing recommendations for intravenous insulin infusions, was implemented at UPMC Harrisburg. Patient data was gathered three months before and after ETIV implementation. Comparative analysis between the pre (47 patients) and post (68 patients) ETIV groups was conducted, with the two groups showing no significant differences in baseline characteristics such as age, sex, BMI, and renal function.
The key results post-ETIV implementation showed a significant reduction in infusion time (37.2 to 35.4 hours, p=0.0003), decreased blood glucose checks (48 to 21.4, p < 0.0001), and a lowered incidence of hypoglycemia (<70mg/dl: 2.26% to 0.35%, p < 0.0001). There were no significant changes in ICU charges ($28,416 to $24,009, p=0.4898), values <40mg/dl (0.03% to 0.00%, p=0.4351), or hospital length of stay (5.1 to 4.5 days, p=0.4993).
The study concludes that the implementation of ETIV led to substantial improvements in clinical outcomes, including reduced hypoglycemia, fewer glucose checks, and shorter infusion time. There was a slight but not statistically significant decrease in length of stay. Thus, ETIV computerized dosing algorithm offers a promising approach in reducing the burden of DKA management and the likelihood of adverse outcomes.
Mathematical Modeling of the Effects of Chronic Kidney Disease and Canagliflozin Treatment on Glomerular Filtration Rate
Andrew Shahidehpour, MEng.; Mudassir Rashid, PhD.; Mahmoud Abdel-Latif, MS; Mohammad Ahmadasas, MS; Mohammad Reza Askari, MS, PhD; Ali Cinar, PhD
Department of Chemical and Biological Engineering, Illinois Institute of Technology Chicago, IL, USA
The management of comorbidities, like chronic kidney disease (CKD), is a major burden associated with type 2 diabetes (T2D) treatment. CKD can complicate blood glucose control and place additional burdens on individuals with T2D. The estimated glomerular filtration rate (eGFR) is one of the key metrics used to track CKD and is expected to decrease along with renal function. Sodium-glucose cotransporter-2 inhibitors (SGLT2Is) have gained popularity as second-like antidiabetic medications that also improve renal and cardiovascular outcomes while slowing the decline of eGFR. The glucose-lowering mechanism of action is limited by eGFR decline and SGLT2Is are contraindicated during the later stages of CKD. Various methods exist for reporting or forecasting eGFR from demographic information and electronic health records. However, there is still a need for flexible and reliable simulation models that incorporate the effects of CKD progression and SGLT2I treatment for capturing eGFR dynamics over time.
A dynamic mathematical model was developed to simulate eGFR progression driven by CKD with and without 100mg or 300mg canagliflozin treatment over 338 weeks. The model parameters were estimated, and results validated, using data from the Canagliflozin Cardiovascular Assessment Study (CANVAS) Program obtained from available publications.
The model fit was evaluated for the placebo group (RMSE: 2.799 mL/min/1.73 m2, MAE: 2.463 mL/min/1.73 m2), treatment group (RMSE: 1.398 mL/min/1.73 m2, MAE: 1.116 mL/min/1.73 m2), and for both groups (RMSE: 2.098 mL/min/1.73 m2, MAE: 1.789 mL/min/1.73 m2).
Mathematical modeling may be used to monitor and forecast CKD progression in T2D. This can aid clinicians and researchers in studying treatment individualization and identifying strategies for long-term T2D care.
Strategies to Aid Human-in-the-Loop in Type-1 Diabetes
Mrunal Sontakke, BS; Faye Cameron, PhD; B Wayne Bequette, PhD
Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute Troy, NY, USA
Individuals with Type-1 diabetes endure the burden of meal announcements for optimal blood glucose control. This work aims to make this task easier for patients through a particle-filter-based meal detection algorithm and a human-machine interface (HMI). The meal detection algorithm extends an existing method that relies on accelerometer data. However, the approach cannot differentiate between the type of meal consumed by the patient, e.g., the motion for drinking water and sugary drink is the same; hence, efforts are made to include blood glucose readings. The HMI is responsible for two tasks: collecting necessary data for the meal detection algorithm and displaying prompts based on output from the algorithm.
An effective probability distribution for meal response is derived from published triple-tracer studies that report meal rate of glucose appearance (RA) curves, which are then scaled based on meal size data from the National Health and Nutrition Examination Survey (NHANES). Similarly, fast-acting insulin tracer studies are used for glucose response curves for insulin. Convolution is utilized to achieve the expected blood glucose distribution based on the combined effects of multiple meals and insulin bolus (Sontakke et al., 2022). These probability distributions are used to detect and identify the meals using observed blood glucose readings from real-world data.
An iPhone and smartwatch app is developed as the HMI. The iPhone app is designed to collect daily patient logs for events affecting their blood glucose and continuous glucose measurements. The smartwatch app collects wrist accelerometer data and periodically transfers data to the iPhone. The app is designed to be intuitive and easy to use without taxing the battery, to maximize data collection. The collected daily logs, blood glucose measurements, and accelerometer data can be used to test the meal detection algorithm.
The work will highlight the results obtained from testing the meal detection algorithm and a comparison of improvement by including the blood glucose data. The algorithm can be extended to include other daily events, such as exercise and sleep, which affect patients' blood glucose levels.
The work will also highlight the HMI based on user input and ready for data collection. Future steps include collecting real-world data from the iPhone app and testing the meal detection algorithm. Once tested, the two can be combined to provide prompts to the user based on the algorithm's output, helping them in meal announcements.
Sontakke M, Cameron F, Bequette BW. Translating Event History to Expected Glucose Distribution. Poster presented at the 2022 Diabetes Technology Meeting (virtual), November 2022.
Impact of a Digital Diabetes Solution on Medication Adherence in Adults in the United States with Type 2 Diabetes Mellitus
Nita Thingalaya, MD, FACP, Dipl.ABOM, CHIE; Darren Frey, PhD; Praveen Potukuchi, PhD; Felix Lee, MPharm, MSc, MBA; David Kerr, MBChB, DM, FRCPE, FRCP
Sanofi US, Bridgewater, New Jersey, United Sates of America
Adherence to a prescribed medication regime is associated with improved outcomes for people living with diabetes. However, impact of digital health technologies on medication adherence (MA) is unclear. Dario Diabetes Solution (DDS) is a digital health technology that combines remote self-monitoring of blood glucose, data visualization, and disease education to facilitate behavior change. Aim of the present study was to determine if DDS impacts MA, and if there is an association between MA and HbA1c in DDS users.
In a retrospective cohort study, adults with T2DM receiving ≥1 diabetes medication, with HbA1c ≥7.0%, and not using CGM (01/01/2017-10/31/2021) were included. MA was measured using proportion of days covered (PDC) over 12-months baseline and 6-month follow-up period. Association of HbA1c change with MA was analyzed by linear regression.
568 DDS users and 1699 non-users participated in the study. Baseline HbA1c levels (mean [SD]) were similar in both groups (users: 9.14 [1.78]; non-users: 9.13 [1.85]). Overall, 78.5% of DDS users and 76.3% of non-users had ≥80% PDC during follow-up (mean difference: 2.20%; p=0.311). Greater improvement in MA was seen among DDS users vs. non-users (10.6% vs. 8.4%; mean difference: 2.16%; p=0.030). Overall, HbA1c in DDS users reduced by 0.28% (p=0.006) more than non-users, with greater HbA1c reduction in the ≥80% vs. ≤60% adherence group (0.36%; p=0.01). HbA1c reduction was greater (0.32%; p=0.04) in DDS users receiving combination treatment than non-users.
In this study, DDS users showed a greater statistically significant improvement in MA and reduction in HbA1c compared to non-users. Additional studies are needed to reinforce that digital health solutions can complement usual care to improve outcomes in diabetes care by supporting MA.
Outcomes of People with Type 1 Diabetes Receiving GLP-1 Receptor Agonists
Tiffany Tian, BA; Rachel Aaron, BA; David Klonoff, MD; C. William Pike, MD; Gavin Hui, MD; Viral N Shah, MD
Diabetes Technology Society Burlingame, CA, USA
To understand outcomes of people with Type 1 Diabetes (PwT1D) in a U.S. population receiving off- label once-weekly glucagon-like peptide-1 receptor agonists (GLP-1RAs).
We reviewed electronic health record (EHR) data from the EVERSANA EHR Dataset, an aggregation of EHR data from >130 million patients in U.S. health systems. We identified PwT1D from 2010-Present with ≥2 instances of T1D diagnosis codes, ≥1 year of continuous insulin use prior to starting a GLP-1RA, and who were receiving weekly albiglutide, dulaglutide, or semaglutide. They were divided into 2 groups: those on GLP-1RAs for >1 year (n=12495) and those on GLP-1RAs ≥1 week and ≤1 year (n=9354). The GLP-1 RA >1yr and GLP-1 RA 1wk-1yr groups were compared using unmatched, basic matching, and propensity score matched (PSM) analyses.
With analysis, the GLP-1RA >1yr group, vs. the GLP-1RA 1wk-1yr group from the start of the GLP-1RA prescription, had a lower hemoglobin A1c (HbA1c) at <3 months (p<0.001), 3-6 months (p<0.001), and 6-12 months (p<0.001). Significantly more PwT1D in the GLP-1RA >1yr group had a HbA1c <7% at 12 months compared to the GLP-1RA 1wk-1yr group (p=<0.001). However, the GLP-1RA >1yr group had a higher BMI at less than 3 months (p=0.0105) and at 3-6 months (p=0.0431) and an insignificantly higher BMI at 6-12 months. The GLP-1RA >1yr group, compared to the GLP-1 RA 1wk-1yr group, had a lower rate of diabetic ketoacidosis per number of matched patients (18/8837 compared to 35/8837, p=0.02).
The results from this real-world data analysis demonstrate potential utility of GLP-1RAs in PwT1D to improve glycemic outcomes. Continued use of GLP-1RAs for >1 year was not associated with increased risk for diabetic ketoacidosis.
Comparison between ID-LC/MS/MS Serum Monosaccharides Method and Enzymatic Methods on Glucose Measurements
Chui Y. Tse, MS; Li Zhang, PhD, Komal Dahya, MS; Fidelia Pokuah, MS; Otoe Sugahara, BS; Uliana Danilenko, PhD; Vincent Delatour, PhD; Hubert W. Vesper, PhD
Centers for Disease Control and Prevention Atlanta, Georgia, USA
Blood glucose is one of the most common clinical tests used for diagnosis and treatment of diabetes, with automated analyzers being the typical instrumentation used in the clinical laboratory. A high throughput ID-LC/MS/MS method that demonstrated good agreement with the CDC Glucose GC-MS RMP was developed for simultaneous quantification of glucose and other free monosaccharides. A comparison study for glucose measurements between the current method and conventional automated analyzer-based enzymatic methods was performed.
Serum samples from 50 individual donors were analyzed in triplicate by seven enzymatic assays from five different manufacturers. The same samples were measured by CDC LC/MS/MS method in duplicate on two different days along with 4 levels of NIST glucose SRM 965b. Glucose concentrations obtained from enzymatic assays were compared with those obtained with our in-house method as the reference.
Allowable bias and total error (TE) limit were evaluated using minimum analytical performances specifications determined based on biological variability.
CDC glucose LC/MS/MS method demonstrated high accuracy with mean bias of -1.04% across 4 levels of NIST SRM 965b. The glucose concentrations of the donor samples, as measured by LC/MS/MS, ranged from 3.02 mmol/L to 13.02 mmol/L (median: 4.70 mmol/L). Other monosaccharides including fructose and mannose were detected in all samples. Using a total error limit of ±9.8%, the number of individual results measured by enzymatic methods falling outside this limit ranged between 2% and 58%.
The current LC/MS/MS assay can be used as a comparative method for evaluating accuracy of glucose measurements. The inconsistencies in the reported glucose values among the enzymatic assays warrants further investigations.
Real-World Glycemic Outcomes Achieved by Three Automated Insulin Delivery (AID) Systems
Robert A. Vigersky, MD; Andrew S. Rhinehart, MD, FACP, FACE, CDCES; Benyamin Grosman, PhD; Anirban Roy, PhD; Louis J. Lintereur, MSc; John Shin, PhD, MBA
Medtronic Diabetes Northridge, California, USA
AID systems employ different control algorithms that may result in differing glycemic outcomes. Since pivotal trials of AID systems have disparate study design, baseline glycemic control, investigators, and participants making direct comparisons difficult, the large number of real-world system users mitigates some of these issues and affords an opportunity to explore algorithmic-driven differences in glycemic outcomes. The present study compared real-world glycemic outcomes data (time in range 70-180 mg/dL [TIR], time below range of <70 mg/dL [TBR], and time above range of >180 mg/dL and >250 mg/dL [TAR]) from three Food and Drug Administration-approved systems (t:slim X2 Control-IQ [CIQ], MiniMed 780G [780G], and Omnipod 5 [OP5]), when used at recommended or default glucose target (GT) or range (GR), with specific active insulin time (AIT) settings.
Real-world data for 780G (Castañeda J. et al. Diabetes Obes Metab. 2022;24:2212-2221), CIQ (Breton M. et al.Diabetes Technol Ther. 2021;23:601-608), and OP5 (Lal R. et al. Diabetes. 2023;72[Supplement 1]:57-OR) were compared. Settings for 780G were GT=100 mg/dL and AIT=2hrs (N=1,262, age >7 years); for CIQ, they were GR=112.5-160 mg/dL and AIT=5hrs (N=7,813, age 6-91 years); and for OP5, they were GT=110 mg/dL and AIT=Any setting (N=5,968, age ≥18 years).
TIR70-180 for 780G, CIQ, and OP5 was 80.7% (mean), 73.5% (median), and 72.1% (mean), respectively. TBR<70 was 2.4%, 0.9%, and 1.5%, respectively. TAR>180 was 13.5%, 19.7%, 26.3%, respectively, while TAR>250 was 2.8%, 4.6%, and not available, respectively.
While definitive conclusions cannot be made about the comparative glycemic benefits of each AID without a head-to-head trial, different algorithmic approaches in AID systems appear to result in diverse real-world glycemic outcomes.
Glucose Sensor-Induced Extracellular Trap Formation Impairs Implant Performance
Kenneth Wood, BA; Jean Gabriel de Souza, PhD; Joseph Cavataio, BS; Tejas Kakunje, BS Ulrike Klueh, PhD
Wayne State University Detroit, MI, USA
Poor biocompatibility and underlying tissue damage caused by glucose sensor (GS) systems could impede further advancement. Extracellular traps (ET) are structures of decondensed DNA decorated with proteins from cytoplasmic granules, which are known to form in response to a variety of stimuli including the elimination of pathogens. Neutrophils, macrophages, mast cells and eosinophils can induce ET formation, called ETosis. However, nothing is known about ETosis in response to device insertion and GS material/coating.
In vitro GS studies include cell viability and ETosis imaging assays using human neutrophils from healthy donors and macrophage cell lines. In vivo studies employed FLOW cytometry to quantify device- induced leukocyte recruitment in a murine air-pouch model. Immunohistopathologic evaluations for ETosis, inflammation, fibrosis and neovascularization at device-sites were conducted in mice and swine.
Our data indicate that tissue responses at sites of GS leads to ETosis, cumulative cell/tissue toxicity, inflammation, and maladaptive wound healing. In vitro studies demonstrated device material-induced cell death and neutrophil extracellular TRAPs (NETosis). Mouse studies demonstrated that influx of inflammatory cells is augmented in the presence GS. NETosis was confirmed in mice and swine. Chronic inflammation, fibrosis, and granulation tissue is seen post 3-day insertion. Cumulative device insertion and device composition (material) are contributors to cell toxicity (cell death, NETosis), and tissue reactions (inflammation, fibrosis and blood vessel regression) at implantation sites.
Future strategies designed to optimize device performance and longevity must mitigate pro-inflammatory factors arising from the device materials and/or insertion site reactions to ensure tissue integrity.
The Proposed Insulin Lispro Biosimilar GN1101DP Shows Pharmacokinetic (PK) and Pharmacodynamic (PD) Bioequivalence to Humalog® in Caucasian and Chinese Healthy Subjects
Eric Zijlstra, PhD; Tim Heise, MD; Grit Andersen, MD; Steffen Selker, PhD; Chun Shen, PhD; Yun Wang
Profil Neuss, Germany
This randomized, double-blind, cross-over study compared PK/PD-properties of Humalog® and GN1101DP, a proposed insulin lispro (INS) biosimilar with an identical primary structure and high physicochemical and bio-functional similarity to Humalog®.
Thirty-five Caucasian and 13 Chinese healthy subjects completed the study and received a single dose of 0.3 U/kg of GN1101DP and Humalog® under euglycemic automated glucose clamp conditions (ClampArt®, plasma glucose (PG) target 81 mg/dL, clamp duration 12 hours post-dose, wash-out period 2-14 days between the two dosings). Plasma INS concentrations were determined by tandem ultra performance liquid chromatography-mass spectrometry.
GN1101DP demonstrated both PK- and PD-bioequivalence to Humalog® with superimposable INS concentration and glucose infusion rate (GIR) profiles. Point estimates (PEs, geometric mean estimates) were close to 100% for the areas under the curve (AUCs) as well as for maximum INS concentrations (Cmax) and maximum glucose infusion rates (GIRmax). The 90% confidence intervals (90% CIs) were within the pre-defined similarity range of 80-125% for all primary endpoints in all and also separately in Caucasian and Chinese subjects (all subjects (n=48): AUCINS.0-12h 99.8 [97.8;101.9] (PE [90% CI]); Cmax 102.3 [97.7;107.2]; AUCGIR.0-12h 98.7 [93.3;104.6]; GIRmax 98.4 [92.4;104.9] // Caucasian subjects (n=35): AUCINS.0-12h 99.5 [96.9;102.3]; Cmax 101.9 96.3;108.0]; AUCGIR.0-12h 96.4 [91.1;102.0]; GIRmax 94.7 [88.2;101.5] // Chinese subjects (n=13): AUCINS.0-12h 100.4 [97.4;103.4]; Cmax 99.5 [94.1;113.2]; AUCGIR.0-12h 106.9[96.5;118.5]; GIRmax 110.4 [99.7;122.3]). Clamp quality was high (mean PG-variations of ~5% and deviations from target <0.6 mg/dL). Both lispro formulations were well tolerated. No injection site reactions occurred.
GN1101DP demonstrated PK- and PD-bioequivalence to Humalog® in Caucasian and Chinese subjects.