2019 Diabetes Technology Meeting Abstracts

doi:10.1177/1932296819897652

. 2020 Mar 2;14(2):361–492. doi: 10.1177/1932296819897652

2019 Diabetes Technology Meeting Abstracts

PMCID: PMC7196873

Abusamaan	Prediction of Iatrogenic Hypoglycemia in the Hospital Using a Machine Learning Algorithm	A1
Abusamaan	Stakeholder Input Regarding Development of a Hypoglycemia Informatics Alert for Hospitalized Patients	A2
Anselmo	Study of Macrophage Inflammatory Responses Caused by Aggregates in Insulin	A3
Arnold	Hand Interface for Collecting Transmission Near Infrared Spectra for Noninvasive Glucose Measurements	A4
Arnold	Impact of Background Spectral Variance on Prediction Accuracy of Noninvasive Near Infrared Sensing of Glucose in People	A5
Arunachalam	Hypoglycemia Rate during Real-World Use of the IQCast Feature in the Guardian™Connect CGM System	A6
Avery	Breakdown of DFU Related Biofilms Using Novel Extremophilic Enzymes	A7
Benesch	New Glucose Clamp Algorithm Improves Clamp Quality with Rapid-Acting Insulins	A8
Biven	Challenges of Statistical Inference Applied to Real World Data in Diabetes Care	A9
Blay	Provider Recommendation of a Mobile Application to Increase Adherence in the Adolescent Patient with Type 1 Diabetes	A10
Brandt	Effects of Sleep on Daytime Glycemic Control in People with Type 1 Diabetes	A11
Brewer	Cause and Effect of Macronutrient Levels on Postprandial Blood Glucose Control	A12
Burgess	Can Severe Hypoglycemia be Eliminated? Introducing Estimated Residual Extracellular (EREI)	A13
Camerlingo	A New Real-Time Algorithm for Preventive Hypotreatments Generation Allows Reducing Frequency and Duration of Hypoglycemia	A14
Cappon	A Non-Linear Bayesian Approach to Personalize Glucose Prediction in Type 1 Diabetes Using a Physiological Model	A15
Carty	Adhesives Enabling Reliable, Multi-Week Wearable Device Attachment	A16
Cembrowski	Determining the Maximum Period of Preanalytical Stability of Glucose in Serial Blood Samples of ICU Patients: A Prerequisite for Data-Mining	A17
Cembrowski	Viewing Obesity-Related Laboratory Tests through the Machine Learning (AI) Lens	A18
Chang	Development of a Low-Cost Hemoglobin A1c Test for the Point-of-Care	A19
Cho	Induced Method by Dynamic Changes of Metabolism for Noninvasive Glucose Measurement	A20
Christiansen	A Phase 3 Comparison of a Ready-to-Use Liquid Glucagon Auto-Injector to GlucaGen® Hypokit® for Severe Hypoglycemia Rescue in Adults with Type 1 Diabetes	A21
Christie	It’s Time for Insulin and Pump-Derived Measure of Engagement: Meal-Time Insulin BOLUS Score Outperforms the Self-Care Inventory	A22
Cichosz	Sleep Duration Predicts Alteration in Metabolic Risk Factors	A23
Collier	Development of a Gestational Diabetes Self-Management and Remote Monitoring Mobile Platform	A24
Delbeck	Insulin Secondary Structure Analysis for Molecular Stability Determination Using Infrared-ATR Spectroscopy	A25
Despa	Technical Development and Clinical Evaluation of a Prototype Integrated Diabetes Management (IDM) System among Insulin-Injecting Patients with Type-2 Diabetes	A26
Diamond	A Control Systems Analysis of “DIY Looping”	A27
Diamond	Assessment of a “DIY Looping” Algorithm Using the UVa/Padova Metabolic Simulator	A28
Diamond	Prediction of Postprandial Glycemic Response from Meal Models of Varying Complexity	A29
Dugas	Contextual Annotations Predict Persistence and Diabetes Outcomes with a Digital Therapeutic	A30
Eichenlaub	Modelling of Glucose Dynamics and Estimation of Insulin Sensitivity from Glucose Data Only	A31
Eichorst	Hospital and Outpatient Insulin Management with a Use of Digital Therapeutics	A32
Ekhlaspour	Outcomes in Pump- and CGM-Naïve Subgroups in the International Diabetes Closed-Loop (iDCL) Trial	A33
Farhy	Clustering and Stratification of CGM Daily Profiles in the International Diabetes Closed-Loop (iDCL) Trial	A34
Franey	Effect of Dawn Phenomenon on Glucose Infusion Rate during Glucose Clamp Studies in Subjects with Type 1 Diabetes	A35
Freckmann	Insulin Pump Accuracy Evaluation – an Update for Bolus Delivery	A36
Freckmann	Post-Market Surveillance of the Accuracy of 18 Blood Glucose Monitoring Systems Available in Europe Based on ISO 15197:2013	A37
Friedman	Light Control of Insulin Delivery: High Insulin Density 2nd Generation Materials	A38
Gal	Glucose Monitoring – What? How? Why?	A39
Gautier	Model Based Assessment of Once Daily iGlarLixi Administration Timing on Glucose Control in Subjects with Type 2 Diabetes: An In-Silico Study	A40
Gill	Effectiveness of Social Media to Disseminate Education about Foot Ulcers among Patients with Type 2 Diabetes Mellitus	A41
Gutierrez-Osuna	Predicting Macro-Nutrients of Foods from Blood Biomarkers	A42
Hardy	A Comparison between Aerobic and Strength Training Exercise on the Reduction of Cardiovascular Disease	A43
He	Impact of Insulin Stability on Intraperitoneal Insulin Catheter Obstruction: A Comprehensive Analysis in Rodents and Type 1 Diabetic Patients	A44
He	Size and Tip Configuration-Dependent Foreign Body Reaction to Intraperitoneal Insulin Catheters	A45
Heise	Investigation of Insulin Formulations by Analytical Methods – Overview and Possibilities for Quality Control	A46
Hernandez	Evaluation of the Analytical Performance of YSI 2900D Compared to YSI 2300 STAT Plus™ for Glucose Concentration Measurements	A47
Hershcovitz	The Effect of Digital Intervention on Glycemic Control in Users with Diabetes	A48
Hilmarsdóttir	A Digital Lifestyle Program to Support Outpatient Treatment of Type 2 Diabetes: A Randomized Controlled Trial	A49
Hobbs	Leveraging an Artificial Pancreas System to Achieve Individual Goals for Management of Type 1 Diabetes	A50
Hu	Monitoring Different Biomarkers in Normal and Diabetic Rats under Stress Using Microdialysis	A51
Hughes	Online Personalization of Hypoglycemia Predictions for People with Type 1 Diabetes	A52
Isganaitis	Closed-Loop Control (CLC) in Teens and Young Adults Improves Glycemic Control: Results from the International Diabetes Closed-Loop (iDCL) Trial	A53
Jensen	Predictors of Post-Prandial Hypoglycemia in Type 1 Diabetes and Continuous Subcutaneous Insulin Infusion	A54
Kastner	“30 vs 90”: The Effect of Angle of Insertion of Insulin Infusion Cannulas on Tissue Histology and Insulin Spread within the Subcutaneous Tissue of Live Swine	A55
Khachaturian	A Non-Invasive Glucose Measurement Device Based on Ultraviolet Light Scattering	A56
Knopp	Adsorption of Insulin to Infusion Lines Is a Function of Flow Rate and is Clinically Significant	A57
Kudva	Control-IQ Users Report High Benefit and Low Burden with System Use during the International Diabetes Closed-Loop (iDCL) Trial	A58
Kuhlenkötter	Control Deviation Is Not Suited as a Clamp Quality Parameter	A59
La Belle	Developments in Insulin Detection Using Antibody and Aptamer	A60
Lal	Fiasp® (Fast-Acting Insulin Aspart) Use with a MedtronicTM 670G System	A61
Lee	Biosensing of Diabetes Markers Using Electroactive Aptamers Based on Square Wave Voltammetry Principle	A62
Legassey	Reducing Size and Power Requirements for Long-Term, Needle-Implantable CGMs	A63
Leon-Vargas	Software Tool for Glucose Monitoring Data Processing in Diabetes Studies	A64
Luxenburg	Use of a Closed Loop Automated Insulin Delivery System in Veterans Over 65 Years	A65
Luzi	Deep Transcranial Magnetic Stimulation for the Treatment of Type 2 Diabetes	A66
Lv	Efficacy and Sensitivity Test of Non-Carbohydrate-Counting Insulin Strategy on the Hybrid Closed Loop System in Type 1 Diabetic Patients: In Silico Results	A67
Mader	The Effect of Glucose Scanning Frequency on Glycemic Control in Individuals with Type 1 Diabetes Using Flash Glucose Monitoring	A68
Mandel	Retrospective Study of Inpatient Diabetes Management Service, Length of Stay and 30-Day Readmission Rate of Patients with Diabetes at a Community Hospital	A69
Manning	Real-World Patient Experience from 5,024 Patients Using the t:slim X2™ Insulin Pump with Basal-IQ® Technology	A70
Melish	Glucose Clearance (mL/min) Assesses Insulin Effectiveness with Usual Mixed Meals Using Continuous Glucose Monitoring (CGM) Data and Body Weight (kg) in a Type 2 Patient on Insulin	A71
Mohebbi	Glycemic Control Assessment Based on Consensus CGM Metrics Using Less than 14 Days	A72
Montaser	SARIMA Local Modelling for Online Glucose Prediction	A73
Mulka	Removal of Phenols from Insulin Suppresses Inflammation & Promotes Blood Glucose Regulation In Vivo	A74
Navarathna	Effect of Using Multiple Sensors on Activity Classification Performance and Smartwatch Battery Life	A75
Newberry	Pilot Study for Non-Invasive Glucose Monitoring by Means of Photoplethysmography	A76
Noaro	Machine Learning Approaches to Enhance Insulin Bolus Dosing in Type 1 Diabetes Therapy	A77
O'Malley	Clinical Management of a Closed-Loop Control (CLC) System: Results from a 6-Month Multicenter Randomized Clinical Trial (RCT)	A78
Oehme	Algorithm Shows Effective Insulin Dosage Calculation in Type 1 Diabetics	A79
Ormsbee	EGP Variation in First 24 Hours of a Critically Ill New Zealand SPRINT Cohort	A80
Owen	Representation of Type 1 Diabetes (T1D) Device Data Using CDISC Clinical Data Standards	A81
Pai	DietHabit: Understanding Dietary Habits in Type 2 Diabetes from Electronic Food Diaries	A82
Patton	Behavior Problems Predict Self-Monitoring Blood Glucose Frequency in Children with Recent-Onset Type 1 Diabetes	A83
Patton	The Mealtime Insulin BOLUS Score Increases Prior to Clinic Visits in Youth with Type 1 Diabetes	A84
Peacock	A Needle Array and Its Application in CGM Sensors	A85
Pfützner	Clinical Evaluation of Acute Interference for a Combined Invasive and Non-Invasive Glucose Meter	A86
Pfützner	De-Escalation Therapy (DET) – An Intensive Short-Term Temporary Pharmacological Intervention to Stop Disease Progression in Patients with Type 2 Diabetes.	A87
Pfützner	Miniaturization of an Osmotic Pressure-Based Glucose Sensor by Means of Nanotechnology Applications	A88
Pfützner	Real World Study for Assessment of the Usability of a Combined Invasive and Non-Invasive Glucose Meter in Patients with Type 1 and Type 2 Diabetes Mellitus	A89
Pleus	Measurement Accuracy of a Newly Developed Prototype System for Non-Invasive Glucose Monitoring	A90
Pleus	Stability of Glucose Concentrations in Frozen Blood Plasma	A91
Pleus	Variation of Measurement Accuracy of Two Continuous Glucose Monitoring Systems during the Day	A92
Portillo	Associations between Demographic Characteristics, Socioeconomic Status, and Glycemic Outcomes in the International Diabetes Closed-Loop (IDCL) Trial	A93
Prendin	A Glucose Specific Metric to Identify CGM-based Predictive Models and Improve the Detection of Hypoglycemic Events	A94
Puleo	Peripheral Focused Ultrasound Stimulation (pFUS): A New Therapy for Metabolic Dysfunction	A95
Pulido	Glucose Clamp Quality Parameters among Study Populations	A96
Pulido	The Role of Continuous Glucose Monitoring in Safety Management of Early Phase Clinical Trials	A97
Rebec	Multi-National Performance Evaluation of the WaveForm Cascade (GlucoMen Day) CGM System	A98
Russell	Performance of the Bihormonal iLet Bionic Pancreas with the Stable Glugagon Analog Dasiglucagon	A99
Russell	Use of the Ultra-Rapid Insulin Fiasp in the iLet Bionic Pancreas	A100
Sainsbury	Predicting Mortality in Both Diabetes Open-Source Clinical Datasets from Free Text Entries Using Machine Learning (Natural Language Processing)	A101
Satyarengga	Preventing Inpatient Hypoglycemia Using Real-Time Continuous Glucose Monitoring: The Glucose Telemetry System	A102
Seley	From Novice to Geek: Teaching Endocrinology Fellows How to Integrate Technology into Practice	A103
Seley	No Type 1 Left Behind: Implementing an Electronic Solution to Missed Basal Insulin Doses During Hospitalization	A104
Sevil	Incorporating the Effects of Psychological Stress Response on Glucose Predictions	A105
Sevil	Incorporating Wearable Physical Activity Trackers to Improve Glucose Predictions for Multivariable Artificial Pancreas Systems	A106
Sheng	Mobile-Enabled Food Logging Is Associated with Improved Glycemic Management in the Real-world	A107
Stoffel	A Novel Method for Determination of Pharmacodynamic Onset of Prandial Insulin Action	A108
Stuhr	Accuracy of the CONTOUR®NEXT ONE Blood Glucose Monitoring System in the Low Blood Glucose Range Using Probability Methodology	A109
Sultan	Past, Present, and Future of Diabetes Technology in Developing Countries	A110
Tolosa	Painless, Bloodless Noninvasive Glucose Monitoring System Using a Fluorescent Glucose Binding Protein	A111
Tsugawa	Glycated Protein Biosensing Based on Engineering Enzymes for 2.5th Generation Amperometric Electrochemical Principle	A112
Uyttendaele	Risk-Based Dosing of Insulin and Nutrition Improves Glycaemic Control Outcomes	A113
Uyttendaele	Women Have Greater (Metabolic) Stress Response than Men	A114
van Baar	Durable Glycemic and Hepatic Improvements After Duodenal Mucosal Resurfacing in Patients with Type 2 Diabetes	A115
Vasiloglou	A Comparison of Macronutrient and Energy Content Estimates by goFOODTM vs Dietitians: A Preliminary Analysis	A116
Vettoretti	Enhancement of the Type 1 Diabetes Patient Decision Simulator to Describe the Behavior of Patients on Multiple Daily Injections	A117
Vettoretti	Modeling the Dexcom G6 CGM Sensor Error	A118
Visentin	In Silico Optimization of Long-Acting Insulin Injection Time in Subjects With Type 1 Diabetes	A119
Waldenmaier	Accuracy of Bolus Delivery of a Novel Patch Pump for Insulin Delivery	A120
Wexler	Blood Glucose Prediction from Pooled Continuous Glucose Monitor Data	A121
Wong	Developing a Branched-Chain Amino Acid Biosensor with Bacterial Periplasmic Binding Proteins for Predicting Patient Risk for Prediabetes and Type 2 Diabetes	A122
Wu	“Smart” Composite Microneedle Patch Delivers Thermostable Glucagon for the Prevention of Nocturnal Hypoglycemia	A123
Zeidan	Clinical Outcomes of V-Go® Wearable Insulin Delivery Device Based on Baseline TDD in Patients with Type 2 Diabetes	A124
Zhang	Clinical Studies of Extended Wear Adhesive Patches for Use on Insulin Infusion Set	A125
Zijlstra	Dance 501 Inhaled Human Insulin: Linear Dose Response in Patients with Type 1 Diabetes	A126
Zijlstra	Faster Absorption and Greater Early Insulin Action of Dance 501 Inhaled Human Insulin vs. Subcutaneous Insulin Lispro in Patients with Type 2 Diabetes	A127

Open in a new tab

Prediction of Iatrogenic Hypoglycemia in the Hospital Using a Machine Learning Algorithm

Mohammed S. Abusamaan, MD, MPH; Mihail Zilbermint, MD; John McGready, ScM, PhD; Shamil Fayzullin; Sam Sokolinsky; Suchi Saria, PhD; Sherita H. Golden, MD, MHS; Nestoras Mathioudakis, MD, MHS

Division of Endocrinology, Diabetes, & Metabolism, Johns Hopkins University School of Medicine Baltimore, MD, USA

mabusamaan@jhmi.edu

Objective:

To develop a prediction model for insulin-related hypoglycemia, defined as a blood glucose (BG) ≤70 mg/dL, occurring within the 24-hours following a given BG reading in hospitalized patients.

Method:

Data from electronic medical records (EMR) of 5 hospitals within the Johns Hopkins Health System were obtained from January, 2015 to July, 2018. Using data from Johns Hopkins Hospital (JHH), the largest hospital in our health system, a prediction model was developed using the random forest classification algorithm. The model was validated internally (by dividing the data set chronologically into training [70%] and testing [30%]) and externally (with the other four hospitals in the health system).

Result:

A total of 1,064,173 BG measurements from 47,217 admissions of 30,514 patients were collected from our 5 hospitals. Using 550,019 BG measurements from JHH, we developed a prediction model containing 39 covariates for the prediction of insulin-related hypoglycemia in a 24-hour horizon. The model achieved C-statistics of 0.80 (95% CI, 0.79-0.80) on internal validation and 0.74 (95% CI,0.73-0.75 ), 0.76 (95% CI, 0.75-0.76), 0.76 (95% CI, 0.75-0.76), and 0.79 (95% CI, 0.79-0.80) on external validation in other hospitals. Using internal validation, the sensitivity, specificity, positive likelihood ratio, and negative likelihood ratios were 80% (95% CI, 0.79-0.81), 79% (95% CI, 0.79-0.79), 3.78 (95% CI, 3.72-3.84), and 0.25 (95% CI, 0.24-0.27), respectively.

Conclusion:

Random forest classification, a machine learning algorithm, can accurately predict incident iatrogenic hypoglycemia using EMR data. Translating this prediction model into a real-time informatics alert has the potential to reduce the incidence of this serious adverse outcome.

Stakeholder Input Regarding Development of a Hypoglycemia Informatics Alert for Hospitalized Patients

Mohammed S. Abusamaan, MD, MPH; Christina T. Yuan, MPH, PhD; Aditya Ashok, MD; Ayman Alam, BS; Sherita Hill Golden, MD, MHS; Nestoras Mathioudakis, MD, MHS

Division of Endocrinology, Diabetes, & Metabolism, Johns Hopkins University School of Medicine

Baltimore, MD, USA

mabusamaan@jhmi.edu

Objective:

To solicit input from key stakeholders regarding development of a real-time informatics alert embedded into the Epic® electronic medical record to predict and prevent insulin-associated hypoglycemia in hospitalized patients.

Method:

A 21-item electronic questionnaire was sent via email from directors of residency and hospitalist programs to their respective trainees/staffs to solicit feedback regarding the importance of iatrogenic hypoglycemia, perceived challenges, perceived utility of an informatics alert, desired format/content of the alert, preferred workflow, and desired accuracy of the alert.

Result:

A total of 102 respondents completed the survey (N=73 physicians, N=23 nurse practitioners, and N=6 physician assistants). All participants agreed or strongly agreed that preventing insulin-related hypoglycemia is important. The majority of respondents (68%) felt that a real-time informatics alert would be a useful tool. The preferred formats for the alert (from most to least desired) were patient header in Epic, text message, glucose/insulin display page, best practice advisory, patient list, or Epic InBasket message. The preferred workflow for the alert was to include recommended action (e.g., insulin dose adjustment) linked to the alert (88% of respondents). Given concern for false positives and alert fatigue, the proportion of respondents who desired an alert specificity of 70-79%, 80-89%, and 90-100% were 30%, 41%, and 20%, respectively. The most common subjective comments were related to alert fatigue, which is commonly encountered by hospital-based providers.

Conclusion:

Stakeholders felt that a hypoglycemia informatics alert would be useful and the preferred format for the alert was a patient header. Alert fatigue was the most cited concern and respondents indicated that alert accuracy was a key consideration. Focus group sessions are underway to further explore the input received in the electronic survey to inform development of the prototype alert.

EE treatment removed significant amounts of biofilm at 55o C. EE treatment removed 41%, 15%, 85%, 65%, 40%, and 44% of Enterococcus faecalis, Klebsiella pneumoniae, Listeria monocytogenes, Pseudomonas aeruginosa 41501, Pseudomonas aeruginosa 15442, Pseudomonas aeruginosa 700888, Proteus mirabilis, or Staphylococcus aureus biofilm(s), respectively. At 41o C, the removal of polymicrobial biofilms was assessed. Treatment with the EE for 4 hrs at 41o C resulted in up to a 52% removal of a polymicrobial biofilm comprised of the previously listed organisms.

Conclusion:

This novel extremophile compound appears able to remove biofilms of clinically relevant organisms. We envision this approach being applied to a DFU via spray application or by powder form. As measurement and management of biofilm becomes more standardized in treatment of DFU, assessing novel means to remove biofilms should become equally standardized. This is especially true in this era of antimicrobial stewardship.

New Glucose Clamp Algorithm Improves Clamp Quality with Rapid-Acting Insulins

Carsten Benesch, PhD; Mareike Kuhlenkötter, MSc; Leszek Nosek, MD; Tim Heise, MD

Profil

Neuss, NRW, Germany

Providers met or exceeded the goals of smartphone app recommendation in the first phase, plateaued in the second phase, and increased in the third phase. All providers reported that the intervention and documentation were easy to utilize but encountered barriers with its use within the organization.

Conclusion:

Provider education on recommendation of a smartphone application increases its utilization in the adolescent population and this is facilitated by a template built into the electronic medical record (EMR). Further research needs to be done on long-term effects of smartphone application recommendations for glycemic control in the adolescent patient.

Effects of Sleep on Daytime Glycemic Control in People with Type 1 Diabetes

Rachel Brandt, BS; Mohammad Askari, MS; Nicole Hobbs, BS; Zacharie Maloney, MS; Ali Cinar, PhD

Illinois Institute of Technology, Department of Biomedical Engineering

Chicago, Illinois, USA

EREI dose adjustment in the eGMS reduced all measures of incidence of hypoglycemia without significantly affecting the time to control or mean BG after reaching goal. In the primary analysis, BG readings less than 40 mg/dl using eGMS without EREI revealed one hypoglycemic event every 129 patient-days of therapy compared to one event every 351 patient-days with EREI. With EREI, the time to control increased from 3.1 to 3.3 hours and mean glucose after first reaching goal increased from 128 to 136 mg/dl.

Conclusion:

EREI integrated into eGMS significantly improves IV-I dosing with significantly reduced incidences of hypoglycemia without significantly affecting glycemic control performance. Analysis revealed severe hypoglycemia (< 40 mg/dl) occurred about a third as often in the EREI adjusted cohort or about once a year of patient therapy.

A New Real-Time Algorithm for Preventive Hypotreatments Generation Allows Reducing Frequency and Duration of Hypoglycemia

Nunzio Camerlingo, MS; Martina Vettoretti, PhD; Giacomo Cappon, MS; Simone Del Favero, PhD; Giovanni Sparacino, PhD; Andrea Facchinetti, PhD

Department of Information Engineering, University of Padova

Padova, Italy

camerlingo@dei.unipd.it

Objective:

American Diabetes Association (ADA) guidelines suggest that subjects with type 1 diabetes (T1D) consume hypotreatments (HTs) whenever hypoglycemia is revealed/detected. However, such a strategy does not avoid the hypoglycemic event. In this work, we propose a new real-time algorithm, using continuous glucose monitoring (CGM) data, with the objective of triggering preventive HTs to reduce both frequency and duration of hypoglycemic events.

Method:

The new algorithm was designed to determine the severity of future hypoglycemia and to trigger preventive HTs by combining CGM values and trend as a measure of risk. The selected comparators were ADA guidelines and a prediction-based algorithm, in which the first HT is triggered based on a 30 min subject-specific prediction. Methods were assessed in-silico using a 7-day realistic simulation involving 100 virtual adults, generated by the UVA/Padova T1D simulator, using as performance metrics time in range (TIR), time in hypoglycemia (THypo), time in hyperglycemia (THyper), and the daily number of HTs (#HTs).

Result:

On median [25th–75th percentiles], ADA guidelines returns 15.1 [13.7–17.4] h/day in target, 22 [8–39] min/day in hypoglycemia and 509 [370–603] min/day in hyperglycemia, with 1.77 [0.69–2.70] HTs/day. Prediction increases TIR (15.5 [14.1–18.0] h/day), reduces THypo (13 [4–27] min/day) and THyper (495 [360–580] min/day), with 1.63 [0.89–2.39] HTs/day. The new algorithm outperforms both previous strategies, providing 15.7 [14.2–18.1] h/day in target, 10 [3–18] min/day in hypoglycemia, and 493 [349–576] min/day in hyperglycemia, also reducing the number of HTs to 1.5 [0.79–2.0]/day.

Conclusion:

The proposed algorithm efficiently generates preventive HTs, increasing TIR, decreasing both THypo and THyper, and, notably, lowering the number of required interventions per day, compared to both ADA guidelines and prediction-based algorithms.

A Non-Linear Bayesian Approach to Personalize Glucose Prediction in Type 1 Diabetes Using a Physiological Model

Giacomo Cappon, MS; Andrea Facchinetti, PhD; Giovanni Sparacino, PhD; Simone Del Favero, PhD

Department of Information Engineering, University of Padova

Padova, Italy

For the GEM, the MAD for intervals of 1 and 2 minutes was 5 mg/dL but increased to 16.4 and 26.9 mg/dL at 5 and 10 minutes, respectively. For the ABL, the MAD remained constant for intervals from 1 minute to 14 minutes [mean= 6.1 ( s=0.52) mg/dL].

Conclusion:

As the two ICU environments are comparable, both located in large tertiary/quaternary academic hospitals, we deem that two intra-patient blood samples can be drawn within 15 minutes and provide roughly comparable glucoses. As the GEM demonstrates a non-constant diurnal variation, the MAD calculation incorporates periods of high variation that complicate glucose comparisons.

Viewing Obesity-related Laboratory Tests through the Machine Learning (AI) Lens

Development of a Gestational Diabetes Self-Management and Remote Monitoring Mobile Platform

Jason Collier, BEng; Jill Fortuin, PhD; Siraaj Adams, MPH

University of Cape Town

Cape Town, Western Province, South Africa

jasoncollier@live.com

Objective:

The impact of gestational diabetes (GD) on maternal and child health has a high morbidity and risk for mortality. The objective is to develop a mobile health (mHealth) platform that optimizes the current GD cycle of care. It is recommended that this platform should include a mobile application (app) for use by patients to self-manage their disease and a web application for use by healthcare professionals to remotely monitor their patients.

Method:

Existing GD management practices and current GD mHealth solutions were researched and the results informed the design of low-fidelity and high-fidelity mock ups of the platform. These mock ups will then undergo usability testing. The insights gained will be used to develop a working prototype of the platform, which will be ready for testing.

Result:

The app has been developed to track the patient’s blood glucose levels via a Bluetooth-enabled glucose meter. The food intake, exercise, and weight gain during pregnancy are manually captured by the patient. The app reminds the patient to take medication, measure glucose levels and attend appointments. A GD educational component is available for the patient throughout the pregnancy. The platform allows for healthcare professionals to remotely monitor and communicate with their patients directly so that they can analyze trends in the data and intervene if necessary.

Conclusion:

The solution benefits patients both financially and time-wise, by reducing the frequency of hospital visits required. It impacts positively on the healthcare professionals by reducing the tediousness of their workload and allowing for remote monitoring of patients. The platform has the potential to optimize the GD management process in South Africa and worldwide.

Insulin Secondary Structure Analysis for Molecular Stability Determination using Infrared-ATR Spectroscopy

Sven Delbeck, MSc; H. Michael Heise, PhD

South-Westphalia University of Applied Sciences, Interdisciplinary Centre for Life Sciences

Iserlohn, NRW, Germany

delbeck.sven@fh-swf.de

Objective:

For the analysis of human and analog insulin formulations and their quality monitoring, reliable analytical methods are required. The protein secondary structure in particular is an important characteristic for determining insulin activity. The molecular structure stability of formulated insulin specimens from pharmacies needs to be investigated, especially after long-term storage under different environmental conditions.

Method:

Different insulin formulations, purchased from pharmacies, as well as pure insulins obtained by ultrafiltration and standard USP insulin in aqueous solution, have been stored under sterile conditions and at different temperatures (0 °C, 20 °C and 37 °C, respectively) using a climatic chamber. For a systematic study of the protein secondary structure over nine weeks, dry film spectra from 1 µl of the insulin solutions have been recorded, using a Bruker ALPHA spectrometer (Ettlingen, Germany), equipped with an attenuated total reflection (ATR) accessory. Analysis of the secondary-structure related infrared bands was carried out by applying various procedures such as band deconvolution, analysis of band shifts or score plots, after principal component analysis of the amide I band interval.

Result:

Small but specific spectral changes in all insulin samples were detected dependent on their storage conditions. These changes correlate well with alterations in their secondary structure and, therefore, with molecular activity. Especially under highest temperature storage conditions, an increased amount of β-sheet fraction could be detected by band deconvolution, underlining the protein structural misfolding.

Conclusion:

While reference analytical methods such as HPLC methods with UV detection or mass spectrometry have been used for the determination of total protein content, FTIR-measurements can be used for monitoring small changes in the secondary structure of insulins leading eventually even to fibril forming.

Technical Development and Clinical Evaluation of a Prototype Integrated Diabetes Management (IDM) System among Insulin-Injecting Patients with Type-2 Diabetes

Mircea Despa, PhD; Jacob Hartsell, BS, MS; Dylan Wilson, BS, MS; Sean Ulrich, BS; Adam Martin, BS; David Bostick, PhD; Tim Hale, PhD; Connor Devoe, BS; Nils Fisher, MPH; Damian Bialonczyk, PharmD; Kamal Jethwani, MD, MPH

BD Technologies Research Triangle Park, NC, USA

Mircea.Despa@bd.com

Objective:

Use of connected tools has been shown to improve glycemic and psychometric parameters among patients with type 2 diabetes (T2D). This research collaboration aimed to develop and pilot a prototype Integrated Diabetes Management (IDM) system for free-living, insulin-injecting, patients.

Method:

The prototype IDM system included: data capture, data algorithms, notification engine, and a clinician portal. Patient data were captured using a prototype insulin pen cap event capture device, a Bluetooth-enabled blood glucose monitor, and a smartphone for step count and meal snapshots. Patients accessed data using a smartphone app, while clinicians reviewed patient data using a clinician portal. Robust design principles and privacy safeguards were applied to develop and test this end-to-end research tool. Clinical assessment of the IDM system consisted of a 6-month, single-arm, open-label pilot study in 35 patients with T2D. Outcomes included adherence to self-care behaviors, app engagement, user satisfaction, usability, and changes in A1c.

Result:

Subjects were 42.9% female, mean age 60.3 (34-75), BMI 33.32 (±10.9) kg/m2 with a baseline A1c of 8.64%. Patient satisfaction and usability were high, with 96.2% and 80.8% of subjects reporting positive ratings, respectively. Average app-use varied over study duration, with 3.00 unique days in week 24 (max:6.19;min:3.00;SD:0.732). Patients with high app use had significant improvement in bolus and basal insulin adherence per week [0.009 p=0.041 (95% CI: 0.0004, 0.018) and 0.016 p=0.000 (95% CI: 0.079, 0.023), respectively]. A1c improved significantly by study end [-0.6% p=0.007, coeff. -0.025 (95% CI:-0.044,-0.007)] No significant improvements were observed in blood glucose adherence, meal snapshots, or absolute step count.

Conclusion:

This research collaboration demonstrates the feasibility of developing and evaluating a prototype IDM system. These findings have informed the development of a consumer mobile application for diabetes management.

A Control Systems Analysis of “DIY Looping”

Travis Diamond, BS; B. Wayne Bequette, PhD; Faye Cameron, PhD

Rensselaer Polytechnic Institute Troy, NY, USA

Artificial pancreas (AP) systems with model-based control algorithms can benefit from using a model for meal glucose absorption. Meal models, which can describe the rate of glucose appearance (RA) from common foods, reject the effects of the meal and provide better blood glucose (BG) control. In this work, multiple meal models of varying complexity were compared based on model fit and predictive performance.

Method:

Five meal models were developed using RA curves from 25 triple-tracer (TT) studies. The least complex model is a linear second-order response approximation, while the most complex uses a multiple model approach. The multiple model approach adapts to the different phases of glucose absorption, and assumes knowledge of time into the meal. Least squares model fits were done on the RA curves. Kalman Filtered predictions of total meal size were performed every 5 minutes using accumulated glucose values from the RA curves.

Result:

The multiple model approach fits the RA data the best, with a 35% improvement over an optimized second-order response model. The average estimated meal size error of predictions made at 20 and 40 minutes into the meal were 13/15/13/12/12 and 13/11/12/11/10 grams, respectively, in increasing model complexity. The multiple model approach improves on the second-order response model by 25% when predicting total meal size throughout the meal. An 8% improvement was seen relative to the nonlinear formulation of the second-order response model.

Conclusion:

A meal model using a multiple model approach was found to fit the TT data best, as well as provide the most accurate predictions of total meal size. This meal model will be used in a multiple model probabilistic framework, which uses time into the meal.

Contextual Annotations Predict Persistence and Diabetes Outcomes with a Digital Therapeutic

Michelle Dugas, PhD; Kenyon Crowley, MSIS, MBA; Weiguang Wang, MS; Anand K. Iyer, PhD, MBA; Malinda M. Peeples, RN, MS, CDE; Mansur Shomali, MD, CM; Guodong (Gordon) Gao, PhD

Center for Health Information and Decision Systems, Robert H. Smith School of Business

University of Maryland, College Park College Park, MD, USA

mdugas@rshsmith.umd.edu

Objective:

Digital therapeutics typically help people manage chronic diseases like diabetes by giving them insights into whatever data they are tracking. Annotations associated with such data may help patients and providers make sense of trends by providing rich contextual information. We explored how patients are using annotations and examined the relationship between annotation and persistent engagement as well as disease outcomes.

Method:

Contextual annotations from 3,142 users of BlueStar® digital therapeutic were analyzed. Annotations could be made in structured (e.g., ‘I feel stressed’) or free-text formats. Thirty-nine percent of users made at least one contextual annotation including a total of 31,422 free-text and 91,572 structured annotations recorded.

Result:

Annotations reflected seven themes of self-management (e.g., diet, medication, mood, and health symptoms). Individuals who recorded annotations persisted in using BlueStar significantly longer than those who did not record annotations, F(3, 3095)= 3.13, p = .02. In a subgroup of users with two self-reported A1C values (n = 378), analysis controlling for demographics and usage showed that the highest note takers exhibited significantly larger declines in A1C than others F(3,357) = 3.55, p =.02. We then developed a measure of patient burden based on annotations. Burden-related annotations were associated with fewer blood glucose readings in range (p = .04).

Conclusion:

A substantial subgroup recorded contextual annotations in BlueStar. The number of annotations was associated with greater persistence and greater A1C improvement, suggesting that annotating may offer unique benefits to digital therapeutic users. In addition, annotations can provide important insight in patient burden, which we found was related to blood glucose control. Results suggest that annotation features may provide powerful insights into patients’ experiences with diabetes while offering benefits to users

Sharoon Gill, MSc, BSc; Kumni Majekodunmi, MD

University of Maryland Baltimore Washington Medical Center Glen Burnie, MD, USA

sgill17@student.umuc.edu

Objective:

The aims of our study were to assess the effectiveness of using social media platforms to improve the efficiency of patient education about Diabetic Foot Ulcers using a structured questionnaire. Social media platforms used for the purpose of this study were Facebook, Twitter, Instagram, Diabetic Forums, and YouTube.

Method:

A structured questionnaire was administered to the outpatients of a preventive care clinic in Glen Burnie, Maryland. All participants were type 2 diabetic. Knowledge regarding diabetic foot ulcers was assessed and scored. After completing the questionnaire, participants watched a YouTube video on diabetic foot care and were provided information on Diabetic Forums, Twitter and Facebook foot care pages. Participants were then asked to engage in any social media platform to learn about foot ulcers.

Result:

Of the 35 individuals who participated in data collection and social media-based health education, only 13 could be contacted after 2 weeks. This was mainly due to participants being unreachable. The percentage of females among those who were reassessed using the same questionnaire was 69%. Of the participants contacted, 46% improved their foot care practices by engaging in a social media platform, 34% did not engage in social media websites but improved their foot care routine after completing the first questionnaire provided, and 20% of participants did not participate in any social media or foot care practices.

Conclusion:

The questionnaire showed the efficacy for using social media websites as a form of patient education. Using social media platforms to disseminate education may be used by clinicians and researchers to provide an efficient method of improving foot care practices among type 2 diabetic patients.

Predicting Macro-nutrients of Foods from Blood Biomarkers

Ricardo Gutierrez-Osuna, PhD

Texas A&M University, Department of Computer Science and Engineering

College Station, TX, USA

rgutier@tamu.edu

Problem Statement:

Measuring food intake is an open problem. Existing methods rely on manual logging, which places a high burden on the participants, or dietary recall, which is costly and unreliable.

Possible Solution:

An alternative approach is to measure the presence or concentration of certain biomarkers in the body. A number of nutritional biomarkers have been identified for certain types of nutrients, such as carbohydrates (e.g., sugars, whole grains), fatty acids, animal protein and various foods/dietary components (e.g., caffeine, citrus). However, most of these biomarkers estimate the average intake of nutrients over extended periods, from weeks to months, but cannot measure food intake continuously (e.g., minutes to hours) or are impractical (e.g., from saliva, urine).

Proposed Innovation:

A more promising alternative is to measure biomarkers in blood or interstitial fluid. by means of implantable or in-dwelling sensors. Such sensors exist to measure glucose, such as the Senseonics implantable or a family of continuous/flash glucose monitors (e.g., Medtronics, Abbott, Dexcom). Glucose correlates strongly with carbohydrate intake but not with other food macronutrients such as fat and protein. As a first step toward development of such sensors, we present a computational study that was used to identify key biomarkers of carbohydrates, fats and proteins. We measured the concentration of glucose, triglycerides and a panel of amino-acids when participants ingested liquid meals with varying but known amounts of macronutrients. Then, we performed feature selection to identify a subset of key biomarkers that are most predictive of the amounts of macronutrients in those meals.

Results:

Results from this computational analysis will be presented.

In this randomized controlled study, patients were consecutively invited to participate. Participants were randomly assigned to an intervention group or a control group. In addition to standard care, intervention group participants used a smartphone application to access the lifestyle program, through which they received personalized recommendations and clinical-guideline-compliant education about healthy lifestyle. Both groups visited the clinic every other month for six months for follow-up measurements, including body weight and blood tests for glycated hemoglobin (A1c) and blood lipids. Furthermore, all participants filled in questionnaires about distress related to diabetes, health-related quality of life, and depression and anxiety. Statistical methods included parametric and nonparametric tests for comparisons both within and between groups.

Result:

A total of 37 patients (23 women) were included, whereof 30 finished, 15 in each group (19% dropout), average age 52.7 ± 10.6 (age range: 25-70) years. No significant differences emerged between the groups, but within the intervention group there was a statistically significant decrease in A1c, from 61 ±21.4 to 52.7 ±15.2 mmol/mol, disease-specific distress (from 19.5 ±16.5 to 11.7 ±13.4) and anxiety symptoms (from 5.4 ±4.0 to 4.1 ±3.8). No significant changes occurred within the control group over the research period. Usage of the app in the intervention group was most frequent during the first months and differed interpersonally.

Conclusion:

Our results indicate that a digital lifestyle program could enhance outpatient treatment outcomes in type 2 diabetes, in terms of both glycemic control and psychological health.

Leveraging an Artificial Pancreas System to Achieve Individual Goals for Management of Type 1 Diabetes

Nicole Hobbs, BS; Iman Hajizadeh, MS; Rachel Brandt, BS; Mert Sevil, MS; Ali Cinar, PhD

Illinois Institute of Technology, Department of Biomedical Engineering

Chicago, IL, USA

nhobbs@hawk.iit.edu

Objective:

Artificial pancreas (AP) systems successfully achieve physician-defined metrics for diabetes management: improving A1c, increasing time in target range, and reducing hypoglycemia. Some individuals with type 1 diabetes (T1D) adore the hybrid closed-loop MiniMed 670G, yet studies presented at conferences (i.e., ENDO2019 and ADA2019) reported that many users stop “automode” within months. The do-it-yourself AP users fine-tune insulin dosing aggressiveness and set-points until satisfied. Their metrics are personal, ranging from: sleeping without continuous glucose monitor (CGM) alarms, exercising without needing supplemental carbohydrates, or being tightly in range even if periodically treating for hypoglycemia. An AP should accommodate a spectrum of preferences.

Method:

A survey collected individual preferences in diabetes management, including daily activities, ideal ranges, insulin dosing and correction doses, and more. Responses from 3 well-controlled individuals were used to adapt the multivariable AP with controller performance assessment (mAP-CPA) to their performance metrics, and the mAP-CPA was tested in 60-day simulations.

Result:

Performance trade-offs exist in managing T1D without manual announcements. Set-point increases of 10 or 20mg/dL led to 9.5% and 19.2% decreases in the number of rescue carbohydrates while increasing daily average glucose by 8.5 and 16.8mg/dL, respectively. Target A1c of 7% (corresponding to 154mg/dL average) was achieved by all 3 set-points. The mAP-CPA can avoid carbohydrate supplementation with modest increases to postprandial glucose excursions and increases to glucose concentrations at exercise onset. These trade-offs are leveraged in the individualized mAP-CPA. Preference of controller aggressiveness, carbohydrate supplementation, and set-point are determined using survey responses and simulations support their feasibility.

Conclusion:

AP settings can be adapted to meet the individual goals of the T1D community through automated, personalized tuning. The inherent trade-offs and personal preferences should dictate how APs handle conflicting issues while maintaining safety.

Monitoring Different Biomarkers in Normal and Diabetic Rats under Stress using Microdialysis

Ping Hu, PhD; Wesia Malik; Diane J. Burgess, PhD

University of Connecticut, Department of Pharmaceutical Sciences

Storrs, CT, USA

ping.hu@uconn.edu.cn

Objective:

Continuous glucose monitoring (CGM) is being used to guide accurate insulin dosing for type 1 diabetes (T1D) patients. In order to obtain a more holistic picture of metabolism and facilitate accurate insulin dosing, additional biomarkers should be identified to help understand the effects of various environmental factors (such as stress) on insulin needs. In addition to glucose, lactate, glycerol, and tissue acidity were measured using tail suspension as a rat stress model.

Method:

Two groups of Sprague Dawley rats (6 normal and 6 diabetic) were used in the study. Microdialysis probes were inserted into rat subcutaneous tissue and left for 30 mins to equilibrate, followed by an 8 min tail suspension and then a 60 min recovery period after suspension. Samples were continuously collected through the microdialysis probes every 4-5 minutes. The pH values of the samples were measured with a pH meter. Glucose and lactate were measured using a YSI biochemical analyzer. Glycerol was measured using a fluorescent assay kit.

Result:

The results showed that tail suspension increased the glucose, lactate, and glycerol levels of both normal and diabetic rats. This effect was much more pronounced for glucose in diabetic rats (nearly 2 times). In addition, the increase in the glucose levels for the normal rats peaked approximately 10 mins after the tail suspension test. The lactate and glycerol levels of normal and diabetic rat increased and peaked approximately 5 mins after the tail suspension test. The pH values of normal rats decreased slightly after suspension, while the pH values of diabetic rats decreased following tail suspension for approximately 15 mins returning to normal values at approximately 30 mins.

Conclusion:

These results suggested that monitoring the trends of multiple biomarkers are useful to validate the onset and duration of stress and may provide useful information to facilitate accurate insulin dosing.

Online Personalization of Hypoglycemia Predictions for People with Type 1 Diabetes

Jonathan Hughes, MS; Marc Breton, PhD

University of Virginia Charlottesville, VA, USA

jh8be@virginia.edu

Objective:

We present and evaluate a system for adaptation/personalization of rule-based hypoglycemia prediction in type 1 diabetes (T1DM). Using both computational and real-world data based simulations, we tested the capability of our system to adapt a rule-based hypoglycemia predictor to specific individuals with T1DM, without prior data collection and/or loss of performance compared to population-based rules.

Method:

An online/stochastic gradient descent procedure was applied to two hypoglycemia prediction algorithms (immediately before exercise and bedtime) based on logistic regression with common T1DM data (insulin pumps, CGMS, etc.) and used to iteratively personalize the regression coefficients from incoming observations. The procedure was tested using computational-numerical simulations and data from clinical studies. Area under the receiver-operator characteristic curves (ROC-AUC) and maximum detection rate at fixed 10% false positive rate (fixed-FPR) served as the metrics of evaluation.

Result:

For real-world data in the 2 hours following exercise onset, population-based rules achieved an ROC-AUC of .6165 vs. 0.7128 for the personalized rule. Corresponding fixed-FPR’s were 0.2257 and 0.2832, respectively. For overnight hypoglycemia detection the results went from 0.7093 to 0.8413, and 0.3208 to 0.5310 in fixed-FPR. Computational-numerical simulations showed significant gains in both ROC-AUC and fixed-FPR over the course of adaptation extending over 100 observations, achieving ROC-AUC within 0.1 of the optimal by 17 observations (median) with median fixed-FPR performance greater than 60% achieved in the same timeframe (vs. 35%).

Conclusion:

Our method was able to achieve personalization of two population-based rules within two weeks with no loss of performance at startup and improved performances in time. This method holds promise for adaptation of most parametrized decision rules, potentially allowing for robust treatment systems adapting to physiological changes.

Closed-Loop Control (CLC) in Teens and Young Adults Improves Glycemic Control: Results from the International Diabetes Closed-Loop (iDCL) Trial

Elvira Isganaitis, MD, MPH; Dan Raghinaru, MS; Louise Ambler-Osborn, NP; Jordan E. Pinsker, MD; Bruce A. Buckingham, MD; R. Paul Wadwa, MD; Laya Ekhlaspour, MD; John W. Lum, MS; Sue A. Brown, MD; Lori M. Laffel, MD, MPH for the iDCL Trial Research Group

Research Division, Joslin Diabetes Center and Department of Pediatrics, Harvard Medical School

Boston, MA, USA

elvira.isganaitis@joslin.harvard.edu

Objective:

Glycemic control in adolescents and young adults with type 1 diabetes (T1D) remains suboptimal. Automated insulin delivery (AID) is a promising approach to improve glycemic outcomes. We assessed eﬃcacy and safety of a closed-loop control (CLC) AID system in teens and young adults.

Method:

We conducted a post-hoc analysis of pediatric outcomes for N=63 youth, age 14 -<25 years (nH, 14 -<18 years; n, 18 -<25 years), in a 6-month multicenter RCT (total N8). Participants were randomized to CLC (Tandem Control IQ™) or sensor augmented pump (SAP, various pumps + Dexcom G6 CGM). Outcomes included time in range 70-180 mg/dL (TIR), A1c, mean glucose, hyperglycemia (>180 mg/dL), and hypoglycemia (<70 mg/dL and <54 mg/dL).

Result:

All 63 pediatric participants completed the RCT. Median T1D duration was 7 years and screening A1c was 8.1%; 37% were pump or CGM naïve at entry. CGM outcomes favored CLC. TIR increased by 13% (3.2 hours/day) with CLC vs. decreasing by 1% (0.2 hours/day) with SAP. Time >180 mg/dL decreased by 12% (2.8 hours/day) with CLC, but increased by 2% (0.4 hours/day) with SAP. Time <70 mg/dL decreased by 1.6% (23 minutes/day) with CLC vs. 0.8% (12 minutes/day) with SAP. CLC use averaged 86% of the time over 6 months. There was one DKA episode in the CLC group and no severe hypoglycemia events in either group.

Conclusion:

In teens and young adults with T1D, CLC use over 6 months was substantial and associated with improved TIR, mean glucose, and time spent in both hyper- and hypoglycemia. Given that youth spend ~½ the day with hyperglycemia, AID systems oﬀer a promising opportunity to improve health outcomes.

Jennifer Knopp, PhD

University of Canterbury Christchurch, Canterbury, New Zealand

jennifer.knopp@canterbury.ac.nz

Objective:

Material adsorption of insulin has been well-observed, resulting in under-delivery of insulin, particularly in the first few hours of infusion where as much as 80% can be lost to adsorption in infusion lines. Adsorption is not widely accounted for in glycemic control in the ICU, and may contribute to poor control and glycemic variability. This study examines material adsorption capacity and its relationship to infusion flow rate.

Method:

Literature data on time-varying insulin adsorption in clinical lines were used to calculate total amounts of insulin adsorbed, indicating material adsorptive capacity (U/m2). These were compared to flow rate to develop a model for material adsorptive capacities.

Result:

Adsorptive capacity decreased hyperbolically with increasing flow rate in both polyethylene (PE) and polyvinyl chloride (PVC) infusion lines. This indicates greater total adsorption to infusion lines in low flow conditions. The proportion of insulin adsorbed vs. delivered is thus higher in low flow and low concentration infusions, which can lose 30-60% of (expected) insulin delivery to adsorption.

Conclusion:

Different materials have different adsorption capacities, and adsorption capacity is a function of flow rate in infusion lines. A relationship was found between total adsorptive capacity of a material with respect to flow, and can be used to predict insulin loss due to adsorption. Adsorptive loss of insulin is greater in low flow and low concentration clinical infusions, such as in neonatal or pediatric intensive care. In these situations adsorption may result in sub-optimal insulin dosing, glycemic variability, and, in worst case, hypoglycemia.

Control-IQ Users Report High Benefit and Low Burden with System Use during the International Diabetes Closed-Loop (iDCL) Trial

Yogish C. Kudva, MD; Dan Raghinaru, MS; Lori M. Laffel MD, MPH; David W. Lam, MD; Carol J. Levy, MD; Laurel Messer, RN, MPH, CDE; Jordan E. Pinsker, MD; John W. Lum, MS; Sue A. Brown, MD; Linda Gonder-Frederick, PhD for the iDCL Trial Research Group

Division of Endocrinology, Diabetes, Metabolism and Nutrition, Department of Internal Medicine, Mayo

Clinic Rochester, MN USA

kudva.yogish@mayo.edu

Objective:

To measure patient-reported technology expectations prior to Control-IQ use and technology acceptance (benefit and burden) and System Usability following Control-IQ use in a large, randomized trial of closed-loop control (CLC).

Method:

In a 6-month randomized, multicenter trial, 168 individuals with type 1 diabetes (14-71 years old, baseline A1c 5.4-10.6%) were assigned 2:1 to CLC (Control-IQ, Tandem Diabetes Care) or sensor-augmented pump (SAP) therapy. At enrollment, 21% of subjects were multiple daily insulin injection users and 79% used an insulin pump; 70% used CGM. Control-IQ subjects answered a Technology Expectations questionnaire at baseline, and a Technology Acceptance questionnaire and System Usability Scale (SUS) at 13 and 26 weeks (all 5-point Likert scales).

Result:

One hundred twelve subjects (age 33±16 years, 48% female) were randomized to Control-IQ, and reported high expectation of benefit (4.0±0.5) and low expectation of barriers (2.0±0.6) at baseline. The Technology Acceptance survey revealed a high perception of benefit at 13 (4.2±0.6) and 26 (4.3±0.6) weeks and low perceived barrier at the same time points (1.8±0.6 and 1.8±0.6). Results from the SUS were similar with high perceived benefit at 13 (4.5±0.5) and 26 (4.5±0.5) weeks and low burden at the same time points (1.7±0.6 and 1.6±0.6), equating to a SUS score of 87 (“excellent” usability).

Conclusion:

The Control-IQ group had high expectations for benefit and low expectation of barriers to the CLC system, which carried through to high perceptions of system usability and technology acceptance after 13 and 26 weeks, indicating high satisfaction with Control-IQ in the first 6 months of use. We are currently exploring associations between patient-reported technology acceptance and glycemic outcomes using Control-IQ.

Control Deviation is not Suited as a Clamp Quality Parameter

Mareike Kuhlenkötter, MSc; Carsten Benesch, PhD; Sascha Heckermann

Profil Neuss, Germany

Mareike.kuhlenkoetter@profil.com

Objective:

The glucose clamp is the gold standard to determine pharmacodynamic characteristics of blood glucose (BG) lowering agents, provided that BG is kept close to the clamp target level (TL). The German Institute of Standardization (DIN) norm for closed-loop controllers (EN 60601-1-10) suggests using control deviation (CD), the mean difference between measured BG and TL, as quality parameter. However, BG fluctuations above and below TL offset one another in CD, potentially leading to wrong conclusions.

Method:

We compared CD and absolute CD (aCD, mean of absolute differences between TL and BG) in a randomized, crossover study. Five healthy subjects received insulin aspart under clamp conditions with ClampArt® using two different clamp algorithms (Biostator and a new PID-algorithm). CD and aCD were calculated over the whole clamp duration of 10 hours post-dose as well as at 2 hour intervals.

Result:

As expected from previous studies, the slow-reacting Biostator algorithm led to BG below TL shortly after insulin dosing and a later “overshoot” with BG above TL. The PID-algorithm substantially reduced these BG fluctuations resulting in lower overall aCD (PID-algorithm: 2.2±0.6mg/dl, Biostator algorithm: 3.6±0.9mg/dl), whereas overall CD even showed lower values for the Biostator algorithm (-1.1±0.2mg/dl vs. -0.6±0.5mg/dl). In most of the shorter time intervals over 2 hours, both CD and aCD supported the notion of better clamp quality for the PID-algorithm without profound differences between algorithms (e.g. first 2 hours, PID- vs. Biostator-algorithm: CD -3.2±1.6mg/dl vs. -5.7±1.3mg/dl; aCD 4.0±1.6mg/dl vs. 5.9±1.2mg/dl).

Conclusion:

CD, proposed as a clamp quality parameter in DIN EN 60601-1-10, can be misleading when positive and negative deviations from TL compensate for each other. We therefore propose using aCD as a quality parameter in glucose clamp studies.

Developments in Insulin Detection using Antibody and Aptamer

Jeffrey T. La Belle, PhD; Curtiss B. Cook, MD; Koji Sode, PhD; Michael R. Caplan, PhD; Blake Morrow, MSE; Connor Beck, BSE

We demonstrated the effectiveness of HF dTMS in the reduction of body weight and in the improvement of metabolic and hormonal parameters related to glucose metabolism in type 2 diabetes with long-lasting effects. The present results constitute proof of principle to consider dTMS a useful tool in the management of type 2 diabetes mellitus.

Efficacy and Sensitivity Test of Non-carbohydrate-counting Insulin Strategy on the Hybrid Closed Loop System in Type 1 Diabetic Patients: In silico Results Dayu Lv, PhD; Leon Farhy, PhD; Marc Breton, PhD

University of Virginia Charlottesville, VA, USA

dl4vf@virginia.edu

Result:

Participants reported high levels of satisfaction (M = 4.49, SD = 0.75) and trust (M = 4.56, SD = 0.72). System usability and utility were also ranked high, with high ratings of “helps me have good blood glucose control” (M = 4.26, SD = 0.86), “helps me feel more in control of my diabetes” (M = 4.31, SD = 0.88), “helps me sleep better at night” (M = 4.09, SD = 0.99) and “helps prevent me from going low” (M = 4.11, SD = 0.92). Patients reported that Basal-IQ is not “complicated to use” (M = 1.70, SD = 0.86) or a “hassle to use” (M = 1.63, SD = 0.88). Patient experience did not significantly vary by age or prior therapy.

Conclusion:

This study demonstrated that patients have positive experiences using Basal-IQ® technology. Future research should continue to examine PROs to determine whether positive patient experience is sustained overtime.

Glucose Clearance (mL/min) Assesses Insulin Effectiveness with Usual Mixed Meals Using Continuous Glucose Monitoring (CGM) Data and Body Weight (kg) in a Type 2 Patient on Insulin

Machine Learning Approaches to Enhance Insulin Bolus Dosing in Type 1 Diabetes Therapy

Giulia Noaro, MSc; Giacomo Cappon, MSc; Simone Del Favero, PhD; Giovanni Sparacino, PhD; Andrea Facchinetti, PhD

Department of Information Engineering, University of Padova

Padova, Italy

noarogiuli@dei.unipd.it

Objective:

In type 1 diabetes (T1D) therapy, the standard formula (SF) used for meal-insulin dosing does not include information on blood glucose (BG) dynamics at meal-time provided by continuous glucose monitoring (CGM) sensors. Thus, the insulin bolus amount can be suboptimal, leading to hyper/hypoglycemic episodes. This work aims to develop and assess in-silico a new formula, extending the feature set and exploiting machine learning methodologies.

Method:

Data from 100 virtual subjects were simulated in single-meal scenario, using the UVa/Padova T1D Simulator, with different pre-prandial conditions in terms of BG and its rate-of-change (ROC). We considered 11 features: carbohydrate intake, insulin-to-carbohydrate ratio, correction-factor, meal-time BG, target BG, insulin-on-board, the SF insulin bolus, basal insulin, body weight, carbohydrates-on-board and ROC. Three models have been developed: least absolute shrinkage and selection operator (LASSO); LASSO trained on a linear basis expansion of the feature set (LASSO-Q); and linear regression model trained using a cost function penalizing insulin-amount overestimations (LR-AS). Performance was assessed on an independent test-set in terms of BG risk index (BGRI) and time in hypo/hyperglycemia (THypo, THyper).

Result:

LASSO, LASSO-Q, and LR-AS lead to better glycemic metrics than SF, by reducing BGRI (8.9 [4.3-15.4], 8.7 [4.2-15.3], 8.5 [4.1-14.9] vs. 9.5 [4.4-17.2]), THypo (0 [0-14], 0 [0-13.3], 0 [0-0.2] vs. 0 [0-28.8] %) without significantly increasing THyper (28.7 [0-38.8], 28.8 [0-38.8], 30 [1.6-40] vs. 27.9 [0.3-37.4] %). Notably, the number of hypoglycemic events is significantly reduced (9827, 9628, 6541 vs. 12929 of SF).

Conclusion:

The proposed machine learning models can potentially enhance the insulin-bolus dosing, suggesting that features on BG dynamics at meal-time, and other easily accessible patient-dependent variables, are key to reduce the post-prandial risk of hypoglycemia.

Clinical Management of a Closed-Loop Control (CLC) System: Results from a 6-Month Multicenter Randomized Clinical Trial (RCT)

Grenye O’Malley, MD; Robert J. Henderson, MS; Carol J. Levy, MD; Jordan E. Pinsker, MD; Gregory P. Forlenza, MD; Laurel Messer, RN, MPH, CDE; Elvira Isganaitis, MD, MPH; Laya Ekhlaspour, MD; John Lum, MS; Sue A. Brown, MD for the iDCL Trial Research Group

Division of Endocrinology, Icahn School of Medicine at Mount Sinai New York City, NY, USA

grenye.o'malley@mountsinai.org

Objective:

Automated insulin delivery is a promising approach to improve glycemic outcomes, but there are limited data on pump setting optimization and clinician input. We examined clinical management of a closed-loop control (CLC) insulin delivery system and its relationship to glycemic outcomes.

Method:

We analyzed personal parameter adjustments in 168 participants in a 6-month multicenter trial. Participants were randomized 2:1 to CLC (Tandem Control-IQ™) vs. sensor-augmented pump (SAP, personal pump+study Dexcom G6™ CGM). The system was initialized with total daily insulin, weight, and insulin pump parameters (BR=basal rates, CF=correction factors, and CR=carbohydrate ratios). Optimization was performed at randomization, 2 weeks, and 13 weeks. Changes were only made at other times for safety issues, participant concerns, or if initiated by participants’ usual care team.

Result:

The CLC group achieved improvements across all primary outcomes compared to SAP. There were 624 encounters for parameter adjustments overall with fewer changes for CLC than SAP (mean 3.5 vs. 4.1/participant). Changes involved BR more often for SAP (CLC 67% vs. SAP 79%), and involved CR (73% vs. 55%) and CF (51% vs. 47%) less often. Changes involved overnight parameter values more often for SAP (63% vs. 75%). Median average CGM time in range (TIR) 70-180 mg/dL during the week preceding and following adjustments increased both for CLC (71.2% to 71.7%) and SAP (61.0% to 63.1%).

Conclusion:

There were fewer parameter adjustments for the CLC group. While SAP benefitted from adjustments, it is less clear if adjustments contributed to the superior TIR achieved by the Control-IQ system. Since Control-IQ can dynamically adjust BR from 0x to 4x the personal parameter, the effect of small changes to any personal parameter is unclear.

Algorithm Shows Effective Insulin Dosage Calculation in Type 1 Diabetics

Markus Oehme, PhD

diafyt MedTech, a pg40 Consulting Group company Leipzig, Saxonia, Germany

markus.oehme@pg40.com

Objective:

Determine the effectiveness of insulin dosage calculation by self-learning systems to optimize glucose levels in type 1 diabetics.

Method:

- CGM & Insulin Management

- Activity Devices/Exercise Monitoring

In addition to the following development topics:

- Diabetic Ketoacidosis Events

- Diabetes History

- Vital Signs and Growth Percentiles

- T1D related Lab Tests

- Metabolic Markers

- Viral Infection History

- T1D Family History

- Islet Autoantibodies

- Genetic Risk

Prior to use, the optical non-invasive device component was calibrated over a period of 3 to 4 days. A comprehensive study with standardized meal experiments at baseline and endpoint and with three months of home use was conducted in patients representing the anticipated user population in the US and EU. For the meal experiments, glucose was assessed at 11 time-points (partly with 2 devices) resulting in a total of 2,729 comparator readings vs. the YSI Stat 2300 reference method.

Result:

At total of 88 patients were included in this analysis (N=43 male, N=45 female, N=24 type 1, N=64 type 2, A1c: 7.4±1.0%; Ethnicities: N=40 Caucasian, N=19 African-American, N=12 Hispanic, and N=14 Asian). For the observed glucose range (53.5-399 mg/dL), a positive NI-mean absolute relative difference (MARD) of 5.1% and a negative NI-MARD of 11.5% were observed (total NI-MARD: 16.6%). In the consensus-error-grid, 99.3 % of the NI-data-points were seen in zones A+B. There were no differences in the meal test results between baseline and endpoint. The NI/INV measurement ratio increased from 1.04 to 2.62 during the observation period. While A1c remained stable (endpoint: 7.5±1.1%, n.s.), there was a substantial reduction by 54% in reported hypoglycemic events (readings<70 mg/dL, p<0.01).

Conclusion:

In conclusion, using the CoG for 3 months at home resulted in stable device performance, an increased measurement frequency, and in a major improvement of glycemic control in patients with type 1 and type 2 diabetes.

Measurement Accuracy of a Newly Developed Prototype System for Non-invasive Glucose Monitoring

Stefan Pleus, MSc; Delia Waldenmaier, MSc; Peter Wintergerst; Nina Jendrike, MD; Manuela Link, MD; Cornelia Haug, MD; Guido Freckmann, MD

Institute for Diabetes Technology, Research and Development mbH at the University of Ulm

Ulm, Germany

stefan.pleus@idt-ulm.de

Objective:

Non-invasive glucose monitoring (NIGM) may be beneficial for people with diabetes in avoiding the need for finger pricking to obtain blood samples. The aim was to assess measurement accuracy of a Raman-based prototype system for NIGM in a proof-of-concept study in a mixed outpatient and in-clinic setting.

Method:

A total of 15 subjects with type 1 diabetes participated in the study which lasted 27 days per subject. Subjects performed standard blood glucose (BG) monitoring with a Contour® next ONE meter and NIGM at the thenar with the prototype system at least 6 times per day. Data from the first 19 to 24 days were used for calibration of the NIGM system. The data from the remaining 3 to 5 days (including 1 in-clinic day each) were used for independent validation of the calibration. In-clinic sessions, during which rapid glucose excursions with high and low glucose values were induced, took place twice (1x on a calibration day and 1x on a validation day). For data from validation days, median absolute relative difference (MedARD) was calculated and Consensus Error Grid (CEG) analysis was performed.

Result:

MedARD was 19.2% for outpatient days and 22.0% for in-clinic days. CEG analysis showed 52.5% and 40.7% of values in clinically acceptable zones A and B, respectively. The remaining values fell within zones C (6.2%) and D (0.5%). No values were found in zone E.

Conclusion:

Although MedARD was comparably high for the newly developed Raman-based prototype, this proof-of-concept study showed promising results. More than 93% of values were found in clinically acceptable zones of the CEG.

Stability of Glucose Concentrations in Frozen Blood plasma

Stefan Pleus, MSc; Annette Baumstark, PhD; Cornelia Haug, MD; Guido Freckmann, MD

Institute for Diabetes Technology, Research and Development mbH at the University of Ulm

Ulm, Germany

stefan.pleus@idt-ulm.de

Objective:

In clinical trials with glucose concentration measurements, the option of storing samples at a study site or at a central laboratory could be helpful. Whereas literature indicates that glycolysis may occur in whole blood samples, data from the literature is inconsistent regarding the stability of glucose in frozen plasma. This study aimed to assess stability of glucose concentrations in separated blood plasma stored at ≤-20 °C.

Method:

Venous blood samples from 21 subjects were collected in lithium-heparin gel tubes, centrifuged, and plasma was aliquoted in 100-µl aliquots. For each subject, glucose concentrations were measured in aliquots immediately after sampling/centrifugation, and after frozen storage for 2, 4, 6, and 8 weeks (i.e. 5 samples per subject). Aliquots were stored for the respective duration in a freezer at ≤-20 °C. Approximately 30 minutes before measurements, aliquots were removed from the freezer for thawing. Because of precipitation in the aliquots, they were re-centrifuged and the liquid supernatant was separated for glucose concentration measurement.

Glucose concentration measurements were performed in duplicate on a hexokinase-based Cobas Integra® 400 plus analyzer (Roche Instrument Center, Switzerland). For each sample, relative differences to baseline (immediate measurement) were assessed. Mean relative differences and 99% confidence intervals (CI) were calculated.

Result:

Relative differences (mean±99% CI) were -0.41%±0.55% (2w), -0.17%±0.42% (4w), -0.89%±0.36% (6w), -0.49%±0.42% (8w), respectively.

Conclusion:

Mean glucose concentration changes did not exceed -0.89% and lower confidence bounds did not exceed -1.25% over up to 8 weeks of frozen storage, indicating stable glucose concentrations.

Variation of Measurement Accuracy of two Continuous Glucose Monitoring Systems during the Day

Stefan Pleus, MSc; Andreas Stuhr, MD; Manuela Link, MD; Jochen Mende, MSc; Cornelia Haug, MD; Guido Freckmann, MD

Institute for Diabetes Technology, Research and Development mbH at the University of Ulm

Ulm, Germany

stefan.pleus@idt-ulm.de

Objective:

The apparent measurement accuracy of continuous glucose monitoring (CGM) systems is influenced by technical aspects, physiologic differences between interstitial and blood glucose (BG), and by daily-life routines like meal consumption. In this study, variation of measurement accuracy during the day was assessed in a well-defined clinical setting.

Method:

A total of 24 subjects with type 1 diabetes participated in this study that lasted for 8 days (7x24 hours) per subject. Subjects each wore a Dexcom G5 (DG5) and a FreeStyle Libre (FL) CGM system in parallel. BG monitoring was performed with Contour next ONE at least once per hour between 06:00 and 22:00, and once during the night at 03:00. DG5 was calibrated twice daily at 07:00 and 19:00 with BG values. Based on BG values and corresponding FL and DG5 values, mean (MARD) and standard deviations (SDARD) of absolute relative differences (ARD) were calculated. Meals were consumed at 08:00, 13:00, and 19:00.

Result:

In this study, MARD and SDARD varied depending on the time of day. Highest MARD values were found approx. 3 hours after breakfast (16.3%). MARD values in the afternoon (12.9% to 15.8%) were also found to be higher than average. Lowest MARD values were found before breakfast for DG5 (8.0%) and before dinner for FL (9.1%). SDARD was highest approx. 3 hours after breakfast and lowest before breakfast.

Conclusion:

In this study, MARD and SDARD values varied depending on the time of day. This study indicates that the apparent measurement accuracy of CGM systems is influenced by study routines and that the impact of daily-life routines should be kept in mind when using CGM in diabetes management.

Associations between Demographic Characteristics, Socioeconomic Status, and Glycemic Outcomes in the International Diabetes Closed-Loop (IDCL) Trial

Katherine Portillo, BA; Christina Tancredi; Laya Ekhlaspour, MD; Bruce Buckingham, MD; Sue A. Brown MD; Boris Kovatchev PhD for the iDCL Trial Research Group

Center for Diabetes Technology Charlottesville, Virginia USA

kap4sp@virginia.edu

Objective:

To determine if certain demographic characteristics were related to improved glycemic outcomes in a study using a mobile closed-loop control (CLC) system.

Method:

This protocol, NCT02985866, is a 3-month parallel group multi-center randomized unblinded trial designed to compare Mobile CLC to sensor augmented pump (SAP) therapy. A total of 125 participants with type 1 diabetes, older than 14 years old, on insulin pump who had A1C < 10.5% completed the protocol. Demographic and continuous glucose monitor (CGM) data were collected during a blinded CGM run-in period, and 3 months of CGM data either on SAP or on CLC using the inControl mobile closed-loop system (Dexcom G4 sensor, Roche Accu Chek insulin pump, and TypeZero Technologies control algorithm running on a smart phone).

Result:

Among subjects aged 14-45 years, there were significant gender differences in baseline glycemic control: mean differences in percent times <54mg/dL (Males↑0.67%), <70mg/dL (Males↑2.31%), and >180mg/dL (Males↓7.92%), and a trending difference (Males↑5.62%) in time in target range (TIR, 70-180mg/dL). CLC equalized men and women with significant improvements from the baseline in all glycemic range outcomes for women. For men, there were significant improvements only in times <54mg/dL and <70mg/dL. Older adults, middle-aged adults, young adults, and adolescents improved TIR by 10.15%, 8.5%, 10.62%, and 7.41%. Individuals with incomes >= $200,000 and < $50,000 improved by 6.12% and 9.77%, respectively. Individuals with no degree, Bachelor’s degree, Master’s degree, and doctoral degree improved by 16.56%, 7.29%, 3.75%, and 13.48% respectively.

Conclusion:

Demographic characteristics are related to differences in glycemic control observed at baseline. While improving overall outcomes, closed loop control appeared to equalize these differences for some parameters, such as gender and education, but gaps persisted among other demographic parameters.

A Glucose Specific Metric to Identify CGM-based Predictive Models and Improve the Detection of Hypoglycemic Events

Durable Glycemic and Hepatic Improvements After Duodenal Mucosal Resurfacing in Patients with Type 2 Diabetes

Annieke CG van Baar, MD; Juan Carlos Lopez-Talavera, MD, PhD; Kelly White, PharmD; Vijeta Bhambhani, MS, MPH; Jacques JGHM Bergman, MD, PhD; Harith Rajagopalan, MD, PhD

Gastroenterology and Hepatology, Amsterdam University Medical Centers, Academic Medical Center

Amsterdam, The Netherlands

a.c.vanbaar@amc.uva.nl

Objective:

Current pharmacotherapies for type 2 diabetes (T2D) require continuous administration and, even with optimal compliance, eventually become insufficient because they do not adequately address pathophysiological defects underlying insulin resistance (IR). High fat/sugar diets alter duodenal hormonal signaling that triggers IR. Duodenal mucosal resurfacing (DMR) is a novel, minimally invasive, outpatient, endoscopic procedure designed to rejuvenate the duodenum lining and restore insulin sensitivity. Herein we report on durability of response through 24 months in patients with T2D who underwent a single DMR procedure in the Revita-1 (R-1) study.

Method:

R-1 was an international, prospective, open-label, multicenter study investigating safety and efficacy of DMR in patients with T2D sub-optimally controlled on ≥1 oral antidiabetic medications for ≥3 months. Eligible patients were aged 28-75 years, body mass index of 24-40kg/m2, and an A1c of 7.5-10.0%. Two-sided paired student’s t-test assessed significance at the 0.05 level.

Result:

Altogether, 34 patients (mean [SD] age and diabetes duration, 56.2 [7.6] and 6.5 [2.4] years, respectively) underwent complete DMR per-protocol (≥9cm circumferential mucosal ablation). At baseline, mean (SD) A1c was 8.5% (0.7). Glycemia indices improved immediately post-DMR, A1c reached 7.5% (0.8) at 6 months and reduction in A1c to 7.5% was maintained through 24 months (N=27, P=0.0020). Most patients (90%) with an A1c improvement from baseline at 12 months maintained improvement at 24 months, with mean reduction of 1.5% at 24 months. Baseline alanine transaminase (ALT) was 38.1 (21.1) U/L; 24 months post-DMR, ALT was 32.5 (22.1) U/L (N=28, P=0.0127). No device-/procedure-related adverse events were noted between 12 and 24 months post-DMR.

Conclusion:

Disease modification in patients with T2D can be achieved with a single DMR procedure as demonstrated by durable glycemic and hepatic improvement through ≥24 months.

A Comparison of Macronutrient and Energy Content Estimates by goFOODTM vs Dietitians: A Preliminary Analysis

Maria F. Vasiloglou, MSc; Ya Lu, MSc; Thomai Stathopoulou, MSc; Lillian F. Pinault, MSc; Stergios Christodoulidis, PhD; Michael P. Jäggi, BSc; Giulia S. Tedde, BSc; Lakshmi G. Singh, PharmD; Elias Spanakis, MD; Stavroula Mougiakakou, PhD

ARTORG Research Center for Biomedical Engineering, University of Bern

Bern, Switzerland

Survival rate to 7 days for the current on-market IIS adhesive patches varied from ~70% to ~90%. The results demonstrated that none of the current infusion set patches are able to provide reliable adherence (with survival rate ≥ 95%) for up to 7 days. The data also indicated that the adhesive materials were key for on-body patch adherence and the hub form factor played an insignificant role in the adhesive patch wear time. Three new adhesive patches were procured and used in the clinical studies. Two of them were able to be worn for 7 days with survival rate ≥ 95% without significantly impacting user experience (i.e., wear comfort or skin reactions).

Conclusion:

Adhesive performance on the body is highly complicated. Although there are many standard tests available, no in- vitro/in-vivo correlation has been established. For the wear performance evaluation of adhesive patches, simulated-use clinical studies proved to be the only valid test method to provide definitive answers to pre-determined, well-defined questions.

Dance 501 Inhaled Human Insulin: Linear Dose Response in Patients with Type 1 Diabetes

Eric Zijlstra, PhD; Oliver Klein, MD; Felix Sievers, MD; Lisa Porter, MD; Blaine Bueche, PhD; Mei-chang Kuo, PhD; Truc Le, MBA; Benjamin J. Stedman, MBA; Melissa Rhodes, PhD, DABT; John S. Patton, PhD

Profil

Neuss, NRW, Germany

eric.zijlstra@profil.com

Objective:

Dance 501 is a novel liquid formulation of human insulin for inhalation (INH) with a small handheld aerosol device. In this randomized, 6-period, cross-over trial, INH vs. subcutaneous (s.c.) insulin lispro (LIS) was investigated during 10-hour glucose clamps to assess the linearity of total and maximum dose-response and dose-exposure of INH in non-smoking subjects with type 1 diabetes (T1D).

Method:

A total of 24 subjects were enrolled (N=9 female, N=15 male; mean ± SD age 44.8 ± 10.2 years; BMI 25.4 ± 2.6 kg/m²). LIS was injected at low (8 U), medium (16 U) and high (32 U) doses. Comparable INH doses were chosen (61.5, 123 and 245.9 U, respectively), assuming 13% relative biopotency (GREL) based on previous studies.

Result:

All subjects completed all dose administrations. Dose linearity of INH was demonstrated for total and maximum activity. Total and maximum exposure of INH increased proportionally from low to medium and from medium to high dose. Mean GREL(%) (CV%) of total pharmacodynamic (PD) activity was slightly lower than assumed (8.9 (56%), 10.4 (42%) and 12.4 (41%) at low, medium and high dose), which must be considered when comparing PD of INH and LIS, as doses were not equivalent. Median onset of action (minutes) was similar for INH vs LIS at low (29min vs 32min), medium (28min vs 29min) and high dose (25min vs 24.5min). Median onset of appearance was 15 minutes for all treatments. No safety issues, no cough, and no changes in FEV1 were observed after dosing.

Conclusion:

Dance 501 showed a rapid onset of activity, linear pharmacokinetic and PD dose response, and good safety and tolerability. INH may, therefore, become a clinically meaningful alternative to rapid-acting injectable insulins.

Faster Absorption and Greater Early Insulin Action of Dance 501 Inhaled Human Insulin vs. s.c. Insulin Lispro in Patients with Type 2 Diabetes

Eric Zijlstra, PhD; Leona Plum-Moerschel, MD, PhD; Marcel Ermer, MD; Oliver Klein, MD; Lisa Porter, MD; Blaine Bueche, PhD; Mei-chang Kuo, PhD; Truc Le, MBA; Benjamin J. Stedman, MBA; Melissa Rhodes, PhD, DABT; John S. Patton, PhD

Profil

Neuss, NRW, Germany

eric.zijlstra@profil.com

Objective:

Dance 501 (INH) is a novel liquid formulation of human insulin for inhalation with a small handheld aerosol device. In this randomized, 6-period, cross-over trial, INH was investigated during 10-hour glucose clamps to assess its dose-response, its time-concentration (pharmacokinetics), and time-action profiles (pharmacodynamics) vs. subcutaneous insulin lispro (LIS).

Method:

A total of 24 subjects with type 2 diabetes participated (N=5 female, N=19 male; mean±SD age 61.8±7.9 years; BMI 30.4±2.6 kg/m², A1c 7.4±0.7%, diabetes duration 10.4±4.8 years, non-smokers). LIS was injected at low (8 U), medium (16 U), and high (32 U) doses; INH was inhaled at equivalent doses, assuming a 13% relative biopotency.

Result:

INH was absorbed faster than LIS (median differences -15 min, p<0.0001), while maximum and total insulin exposure were similar (p>0.3) for all dose comparisons. The faster kinetics resulted in more rapid onset of action (median differences -6.5 to -20 min, p<0.02) and greater action in the first hour after administration (median relative differences 45 to 107%, p<0.05). INH and LIS demonstrated a linear dose-response relationship and comparable total pharmacodynamic action (p>0.2 for comparison at each dose level). Median relative bioavailability of INH was 11.9, 13.0 and 13.8% at low-medium-high dose levels, respectively and biopotency of INH was 12.3, 13.0 and 12.2%, respectively. No safety issues, no cough, and no acute changes in spirometry were observed with any inhalation.

Conclusion:

Dance 501 showed faster insulin absorption, more rapid onset of action, and greater early insulin action vs. LIS. From 1 hour post-dosing, insulin exposure and action were comparable between treatments. Inhalations were well tolerated and without cough. Dance 501 may become a valuable and clinically meaningful option for insulin treatment in patients with type 2 diabetes.

PERMALINK

2019 Diabetes Technology Meeting Abstracts

Prediction of Iatrogenic Hypoglycemia in the Hospital Using a Machine Learning Algorithm

Objective:

Method:

Result:

Conclusion:

Stakeholder Input Regarding Development of a Hypoglycemia Informatics Alert for Hospitalized Patients

Objective:

Method:

Result:

Conclusion:

Study of Macrophage Inflammatory Responses Caused by Aggregates in Insulin

Objective:

Method:

Result:

Conclusion:

Hand Interface for Collecting Transmission Near Infrared Spectra for Noninvasive Glucose Measurements

Objective:

Method:

Result:

Conclusion:

Impact of Background Spectral Variance on Prediction Accuracy of Noninvasive Near Infrared Sensing of Glucose in People

Objective:

Method:

Result:

Conclusion:

Hypoglycemia Rate during Real-world use of the IQCast Feature in the Guardian™ Connect CGM system

Objective:

Method:

Result:

Conclusion:

Breakdown of DFU Related Biofilms Using Novel Extremophilic Enzymes

Objective:

Method:

Result:

Conclusion:

New Glucose Clamp Algorithm Improves Clamp Quality with Rapid-Acting Insulins

Objective:

Method:

Result:

Conclusion:

Challenges of Statistical Inference Applied to Real World Data in Diabetes Care

Objective:

Method:

Result:

Conclusion:

Provider Recommendation of a Mobile Application to Increase Adherence in the Adolescent Patient with Type 1 Diabetes

Objective:

Method:

Result:

Conclusion:

Effects of Sleep on Daytime Glycemic Control in People with Type 1 Diabetes

Objective:

Method:

Result:

Conclusion:

Cause and Effect of Macronutrient Levels on Postprandial Blood Glucose Control

Objective:

Method:

Result:

Conclusion:

Can Severe Hypoglycemia be Eliminated? Introducing Estimated Residual Extracellular (EREI)

Objective:

Method:

Result:

Conclusion:

A New Real-Time Algorithm for Preventive Hypotreatments Generation Allows Reducing Frequency and Duration of Hypoglycemia

Objective:

Method:

Result:

Conclusion:

A Non-Linear Bayesian Approach to Personalize Glucose Prediction in Type 1 Diabetes Using a Physiological Model

Objective:

Method:

Result:

Conclusion:

Adhesives Enabling Reliable, Multi-Week Wearable Device Attachment

Objective:

Method: