Abstract
Prediction of Iatrogenic Hypoglycemia in the Hospital Using a Machine Learning Algorithm
Division of Endocrinology, Diabetes, & Metabolism, Johns Hopkins University School of Medicine Baltimore, MD, USA
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
Division of Endocrinology, Diabetes, & Metabolism, Johns Hopkins University School of Medicine
Baltimore, MD, USA
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.
Study of Macrophage Inflammatory Responses Caused by Aggregates in Insulin
Medtronic Diabetes Northridge, CA, USA
Objective:
Continuous subcutaneous insulin infusion (CSII) using an insulin pump has been available as an insulin delivery method for over 30 years. The infusion set is a conduit that delivers insulin from the pump to the patient’s body via the infusion site. Current infusion sets are labeled to be replaced every 2 to 3 days, due to various site-related complications. One of the significant complications is the decreased insulin absorption, often attributed to the layer of inflammatory tissue which may function as a mechanical barrier to insulin flow into adjacent vascular tissue. Monocytes and resident tissue macrophages are key regulators of tissue inflammation. The objective of this study was to investigate the inflammatory responses triggered by macrophage cells reacting to insulin aggregates delivered into the subcutaneous site.
Method:
Murine macrophage cells were used in the study. The cells were cultured in well plates and exposed to various chemical stimulants such as insulin solutions infused through various pump/infusion set use conditions (generating various amounts of aggregates/particles) in combination with infusion site related biomolecules (e.g. fibrinogen). Macrophage cell growth and morphology were observed. Cytokine quantifications by ELISA were conducted on the cell culture supernatants to evaluate macrophage inflammatory responses to stimulants.
Result:
More macrophage growth was observed when the cells were exposed to the combination of fibrinogen and insulin solution with elevated levels of aggregates. Also, a few pro-inflammatory cytokines increased significantly in the corresponding cell culture supernatants.
Conclusion:
Macrophage inflammatory responses were significantly impacted by the trace aggregates/particles in the insulin solution. Therefore, maintaining insulin formulation stability in the insulin pump/infusion set system is critical to mitigate the inflammatory responses at the subcutaneous infusion site.
Hand Interface for Collecting Transmission Near Infrared Spectra for Noninvasive Glucose Measurements
University of Iowa Department of Chemistry
Iowa City, IA, USA
Objective:
Direct noninvasive in vivo measurements of glucose demand an analytical signal that originates from the glucose molecule, as opposed to indirect methods that are based on secondary, correlative information. The objective of this study is to evaluate the clinical utility of a direct spectroscopic measurement of glucose in skin by using an innovative hand interface coupled with net analyte signal (NAS) calibration methods.
Method:
The hand interface was constructed such that skin on the back of the hand was held within a conical shaped chamber. Two sapphire rods were aligned 180° across this skin sample so the incident radiation was provided by one and radiation transmitted through the skin was collected by the other. This interface was used to collect near infrared spectra over 4200-4800 cm-1 during a series of glucose tolerance tests (GTT) performed with a cohort of subjects with type 1 diabetes.
Result:
Data from each GTT were analyzed by an NAS calibration method that generates a calibration vector originating from the pure component spectrum of glucose. This analysis used a set of spectra collected under an initial fasting phase before the subject ingested a meal. The NAS derived concentrations of glucose were compared to results from a standard glucose meter. Collectively, 153 measurements were obtained over 15 separate GTT sessions performed with five different people. Results include a Mean Absolute Relative Difference (MARD) of 12.8% combined with an R2 of 0.81 and 81.0% of observations within the A-region of the Clarke Error Grid.
Conclusion:
These results demonstrate the potential of noninvasive glucose measurements from an interface that enables the collection of near infrared spectral transmission across human skin. The NAS approach adds to the significance of these findings by reducing the impact of correlative effects.
Impact of Background Spectral Variance on Prediction Accuracy of Noninvasive Near Infrared Sensing of Glucose in People
University of Iowa Department of Chemistry
Iowa City, IA, USA
Objective:
Glucose measurements in people using near infrared spectroscopy demand a unique spectral signature for glucose relative to the spectral features associated with the skin matrix. Variations in skin spectra represent a long-standing challenge for selective noninvasive measurements in people. The objective of this study was to characterize variations in background spectra for a series of repetitive noninvasive spectra collected over a period of several hours from a cohort of volunteers.
Method:
For each subject, a series of diffuse reflectance spectra was collected over the 4250-4850 cm-1 spectral range. Initially, fasting spectra were collected followed by ingestion of a meal composed of 75 grams of carbohydrates. Capillary blood glucose measurements were made periodically throughout. In total, 11 profiles were collected from three volunteers, each with type 1 diabetes.
Result:
Overall, spectral signal-to-noise ratios ranged from 11,000 to 14,000 for the ≈30 second integration period used in collecting these noninvasive skin spectra. Partial least square (PLS) calibration models were generated by using the first 80% of the collected spectra of each profile to generate a calibration model and this model was used to predict the concentration of glucose for the last 20%. For most profiles, major deviations were observed when comparing glucose concentrations obtained during the calibration and prediction periods. Similarly, net analyte signal (NAS) calibration models were created for each profile and prediction accuracy degraded significantly outside the calibration period.
Conclusion:
Accurate glucose concentration measurements demand proper characterization of the background spectral variance, regardless of the multivariate method (i.e., PLS or NAS). The complexity and variability of the background skin spectra are shown to represent the principal challenge for successful noninvasive glucose measurements for people with diabetes.
Hypoglycemia Rate during Real-world use of the IQCast Feature in the Guardian™ Connect CGM system
Medtronic Northridge, California, USA
Objective:
The Sugar.IQ™ diabetes assistant mobile application of the Guardian™ Connect CGM system provides the IQCast feature that generates a probability score (low, medium or high) representing the likelihood of low glucose episodes within a 1- to 4-hour window. The user can track this information and receive a notification when a high probability score is predicted. The objective of this study is to analyze glycemic outcomes before and after IQCast use.
Method:
Deidentified data from 337 individuals living with diabetes and who used the Sugar.IQ™ diabetes assistant prior to the introduction of the IQCast feature were selected for analysis. Mean hypoglycemic (<70mg/dL and <54mg/dL) episode frequency and time below range (TBR) were compared before (N=19,517 days of data) and after (N=29,098 days of data) delivery of the first notification from IQCast.
Result:
There was an average 3.6 IQCast notifications/day. After IQCast introduction, the overall and nighttime mean±SD (median with CI’s) frequency of hypoglycemic episodes (<70mg/dL) decreased from 16.5±15.3 (12.1, 13.9-17.2) episodes/month to 13.8±12.9 (9.8, 11.5-14.3) episodes/month and from 5.0±4.5 (3.8, 4.1-5.3) episodes/month to 4.0± 4.4 (2.7, 3.3-4.5) episodes/month, respectively (p<0.001 for both). The overall and nighttime mean frequency of severe hypoglycemic episodes (<54mg/dL) decreased from 4.1±6.3 (2.2, 3.3-4.7) episodes/month to 3.4±4.8 (1.7, 2.7-3.8) episodes/month (p<0.001) and from 1.5±2.2 (0.6, 1.0-1.6) episodes/month to 1.1± 2.1 (0.5, 0.8-1.4) episodes/month (p<0.05), respectively. IQCast improved (decreased by ≥1 episode/month) hypoglycemic episode frequency for 52% of the users and kept the rate unchanged for 20% of the users. Mean TBR (<70mg/dL) and TBR (<54mg/dL) decreased from 3.4±3.6% (2.5, 3.1-3.9) to 3.2±3.1% (2.1, 2.7-3.4) and from 0.7±1.2% (0.3, 0.6-0.9) to 0.6±0.9% (0.3, 0.5-0.7), respectively (p<0.001 for both).
Conclusion:
Results from this study demonstrate that using IQCast technology decreases the frequency of both overall and nighttime hypoglycemic episodes.
Breakdown of DFU Related Biofilms Using Novel Extremophilic Enzymes
Liberty Biosecurity Arlington, VA, USA
Objective:
There is emerging evidence supporting the claim that bacterial biofilms influence the healing rates of diabetic foot ulcers (DFU). The extracellular polymeric substances (EPS) that make up biofilms are generally carbohydrates, DNA/RNA, proteins, lipids, and cellular debris. The goal of this research is to evaluate the effectiveness of extremophilic enzymes (EE), purified from a novel organism, to hydrolyze bacterial biofilms and aid in their removal.
Method:
Biofilm removal efficacy of the novel EE was determined against medically relevant biofilms grown under static growth conditions. For static growth, a 96-well PVC microtiter plate assay was used. Following biofilm growth, wells were treated with the EE cocktail and stained with crystal violet to quantify any remaining biofilm. The EE treatment was 4 hours at 55o C or 41o C.
Result:
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
Profil
Neuss, NRW, Germany
Objective:
Glucose infusion rates in automated glucose clamps are determined by an appropriate algorithm that aims at keeping blood glucose (BG) close to a predefined target level. This automatization allows bias-free determination of the pharmacodynamics of BG-lowering agents. However, some algorithms, such as the frequently used Biostator algorithm, struggle to keep BG close to target in cases of rapid changes in the insulin effect. We therefore developed a new Clamp algorithm based on a proportional-integral-derivative controller (Clamp-PID algorithm) and compared the clamp quality achieved with the Biostator algorithm.
Method:
Numerical simulations were used to develop and optimize the new algorithm to minimize BG fluctuations. After in-vitro validation studies, the two algorithms were compared in a small, single center, randomized, crossover euglycemic glucose clamp study in healthy subjects using insulin aspart (n=5) or insulin glargine (n=3). Glucose clamp quality parameters consisted of precision (standard deviation of all BG values) and absolute control deviation (aCD: mean of absolute differences between target level and BG values).
Result:
The new Clamp-PID algorithm reduced glucose fluctuations in the insulin aspart clamps by nearly 40% (precision: 4.0±1.1% (mean±SD) vs. 6.5%±1.3% with the Biostator algorithm) and kept BG closer to target (aCD: 2.2±0.6mg/dl vs. 3.6±0.9mg/dl). In contrast, the performance of both algorithms was very similar in the insulin glargine clamps (precision: 3.2±1.0% vs. 3.1±0.8%, aCD: 1.7±0.5mg/dl vs. 1.7±0.5mg/dl).
Conclusion:
The new Clamp-PID algorithm significantly improves glucose clamp quality with rapid-acting insulins and shows similar clamp quality for basal insulins. The new algorithm will now be implemented in the ClampArt device.
Challenges of Statistical Inference Applied to Real World Data in Diabetes Care
mySugr GmbH Vienna, Austria
Objective:
Assessing diabetes therapy performance from Real World Data (RWD) is difficult as the data quantity of SMBG measurements shows strong variability. Meeting the measurement regimens used in clinical setups is difficult for the majority of patients and potentially leads to a strong selection bias. Analyses for SMBG constraints applicable to RWD are missing but seem necessary given the increasing interest in RWD studies and the lack of standardization.
Method:
We found that previous studies using SMBG-based metrics often lack a predictive performance assessment (with respect to SMBG frequency) with Average Daily Risk Range (ADRR) being an exception (3 SMBG/14days). We see two potential risks when applying no or un-validated constraints. First, using loose inclusion criteria, such as mean frequency, one might falsely include patients with skewed SMBGs. Second, hard constraints might favor highly motivated users. In order to leverage these biases, we formalized flexible inclusion criteria.
Result:
Gn/Nkn out of N days:SMBG(k) formalizes data constraints as an observational period of N days in which a subset of n days is required to each contain ≥ k blood glucose readings. The formalism can easily be applied to published ADRR constraints (i.e., 3 SMBG/day in 14 out of 30 days) as G314/30. Our previous work modeled hypoglycemic excursion probabilities using kernel density functions and found G414/30 using continuous glucose monitoring (CGM)-based predictive performance assessment.
Conclusion:
The proposed G-classification provides flexible criteria for RWD metric assessments while limiting inclusion biases. Applied to glucose control metrics, it can help to ensure data quality in RWD centric studies. We are planning further performance validation for other metrics such as eA1c and CV.
Provider Recommendation of a Mobile Application to Increase Adherence in the Adolescent Patient with Type 1 Diabetes
University of Maryland School of Nursing Baltimore, MD, USA
Objective:
The aim of this quality improvement project was to examine the mobile applications that have worked to improve self-care in the adolescent patient and give providers the knowledge they need to feel comfortable recommending a mobile app to their patients as part of routine visits at the diabetes clinic.
Method:
Inclusion criteria for provider recommendation of a mobile app for purposes of this project included all patients aged 14-17 years seen in the three-month period from September to December, a total sample size of N=78 patients. A ‘smart phrase’ was created for documentation in the electronic medical record and providers were given handouts and educated on the process of recommendation. A run chart and descriptive statistics were used to analyze the data during and after the interventions.
Result:
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
Illinois Institute of Technology, Department of Biomedical Engineering
Chicago, Illinois, USA
Objective:
There is a complex relationship between sleep and diabetes. Most of the reported research has focused on insulin resistance and type 2 diabetes. The relationship between sleep and daytime glycemic control in people with type 1 diabetes (T1D) must be better understood in order to incorporate new modules into a multivariable artificial pancreas for better glycemic regulation.
Method:
Subjects with T1D, ages 18 to 60 years, were each monitored during weekdays for three weeks. Each participant wore an at-home automatic sleep-staging device and a continuous glucose monitor (CGM). Participants maintained a constant activity schedule and meal compositions during the three weeks. Quantitative descriptive features including total daily dose, mean amplitude of glycemic excursions, and coefficient of variation were developed with the daytime insulin and CGM data along with corresponding sleep data from the previous night. K-means clustering was used to determine relationships between sleep characteristics and daytime glycemic control. Analysis was done across the entire group and for each individual participant.
Result:
When analyzed together, sleep characteristics did not show a common effect on daytime glycemic control. However, when analyzed individually, participants all had distinct clusters of data that showed that sleep influenced their next day’s glycemic regulation. Furthermore, most individuals exhibited unique relationships based on differing sleep characteristics including measures such as sleep efficiency, percent of the night in deep sleep, and total sleep time.
Conclusion:
These results suggest that sleep has distinct, individualized effects in T1D. There is a need for personalized sleep effect models to have successful predictions for aiding in daytime glycemic control. Larger population studies on the effects of sleep and daytime glycemic control are needed to find if there is a population effect in addition to individualized effects.
Cause and Effect of Macronutrient Levels on Postprandial Blood Glucose Control
Glenelg High School Glenelg, MD, USA
Objective:
Controlling blood glucose variation and specifically minimizing hyperglycemia are important to prevent the development of diabetic complications. Macronutrient levels of meals have an effect on postprandial blood glucose. This study evaluated how to maintain postprandial blood glucose within the desired range of 70-180 mg/dL by manipulating macronutrient levels.
Method:
One teenage male, type 1 diabetic subject developed meal plans for three meals per day by setting macronutrient levels to different mass ratios. Postprandial blood glucose results, measured by a continuous glucose monitor (CGM) for six days, were used to assess the effect of various macronutrient ratios.
Result:
Analysis of meal macronutrient levels versus CGM plots showed that when the mass ratio of carbohydrates to protein, compared to the mass ratio of carbohydrates to fat, was maintained at a value of approximately 1.1, the blood glucose after meal consumption showed lower glycemic excursions and remained within the desired range of 70-180 mg/dL. If the ratios were either less than 0.6 or greater than 1.6, then the postprandial blood glucose showed a significant rise and remained high, often needing additional insulin to restore blood glucose to the acceptable range.
Conclusion:
Ratios of carbohydrates to proteins and fats have an impact on postprandial blood glucose excursions. Meals can be formulated by setting the desired macronutrient ratios in order to obtain postprandial blood glucose levels within the desired range, thereby minimizing hyperglycemia and the risk of diabetic complications. The results are specific for this subject and may vary individually. Any patient seeking to maintain glucose homeostasis can easily a similar study.
Can Severe Hypoglycemia be Eliminated? Introducing Estimated Residual Extracellular (EREI)
Monarch Medical Tech., LLC Charlotte, NC, USA
Objective:
Prevention of hypoglycemia is a goal of electronic glucose management systems (eGMS). Estimated Residual Extracellular Insulin (EREI) was introduced as an adjustment for future intravenous insulin (IV-I) activity. Designed to reduce hypoglycemia, performance of EREI adjustments is reported in this retrospective analysis of hypoglycemia in hospital systems.
Method:
Retrospective data from 84 community hospitals were stratified to two cohorts related to date of introduction of EREI adjustments in the eGMS. The primary analysis is a comparison of incidence of hypoglycemia in the two cohorts for subjects with 140 mg/dl as the upper glucose goal. Secondary analyses include time to control, mean blood glucose (BG), and variance after first reaching the goal of 140 mg/dl, and the primary analysis applied to all goals. No data exclusions were applied, including late glucose determinations which are commonly associated with hypoglycemia.
Result:
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
Department of Information Engineering, University of Padova
Padova, Italy
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
Department of Information Engineering, University of Padova
Padova, Italy
Objective:
Our aim is to achieve an accurate prediction of glucose concentration in type 1 diabetes (T1D) and to improve glycemic control by enabling proactive treatment of hypo-/hyperglycaemia. To do so, we leveraged a physiological nonlinear model that described the glucose-insulin dynamics in T1D. The model is personalized for each patient to account for the large intra-individual variability.
Method:
We started from the UVa/Padova T1D Simulator mathematical model. This large model was simplified to obtain a reduced nonlinear physiological model that, compared to the original model, had few parameters but still accurately described glucose-insulin dynamics. Since the simplified model was not identifiable, its parameters were identified using patient data by adopting a Bayesian approach, specifically Markov Chain Monte Carlo, which exploits a priori information on parameter distributions to deal with identifiability issues. Finally, a particle filter has been implemented obtaining glucose predictions at different prediction horizons (PH). This method has been tested using the UVa/Padova T1D Simulator, generating in silico data for 100 virtual subjects over 7 days with multiple meal scenarios. The first 3 days were used for model parameter estimation. Prediction accuracy were evaluated using the remaining 4 days of data in terms of root mean square error (RMSE) and coefficient of determination (COD).
Result:
The methodology achieves good prediction accuracy both in the short and the long term. In particular, median RMSE and COD are 11.20 mg/dL and 90.97 %, respectively, for PH = 30 min, and 16.99 mg/dL and 78.40 %, respectively, for PH = 180 min.
Conclusion:
The proposed approach reliably predicts future glucose concentration both in the short and the long term. Future work will assess this methodology using retrospective clinical data.
Adhesives Enabling Reliable, Multi-Week Wearable Device Attachment
Avery Dennison Medical Technologies Chicago, IL, USA
Objective:
Innovations in diabetes care, like the advent of convenient, low-profile continuous glucose monitors (CGMs), are beginning to outpace advances in the adhesive materials that are needed to interface the device with the body. The promise of sophisticated sensing and/or drug delivery devices cannot be realized without the means to hold them on the skin for weeks at a time. Our objective was to engineer a state-of-the-art adhesive system that offers reliable, multi-week wear times.
Method:
Sham devices consisting of a plastic disk bonded to a skin adhesive tape in a skirt-style construction were assembled using three different adhesive material combinations. With informed consent, healthy human volunteers were recruited to wear a randomized selection of prototypes on their abdomens and arms, recording the time of application and the failure time when the prototype eventually detached. Instances of skin irritation and discomfort during wear were also noted throughout the study. The study lasted a total of 30 days. Reliability was assessed by analyzing the time to adhesive failure using Kaplan-Meier non-parametric survival analysis.
Result:
All three prototypes exhibited excellent reliability and were well-tolerated with minimal irritation or discomfort during wear. The best-performing prototype exhibited 100% reliability within the first ten days of wear and 86% reliability after 21 days of wear time. The mean time to failure was 22 days.
Conclusion:
With the right design, adhesive tapes can provide multi-week wear times. These materials open the door for innovative diabetes care technologies that are more convenient, comfortable, and economical for the patients who use them.
Determining the Maximum Period of Preanalytical Stability of Glucose in Serial Blood Samples of ICU Patients: A Prerequisite for Data-mining
Laboratory Medicine and Pathology, University of Alberta Edmonton, Alberta, Canada
Objective:
The accuracy of blood gas analyzer (BGA) glucose can be evaluated by comparing point-of-care intensive care unit (ICU) patient BGA glucose to glucose measured in the central laboratory (CL) on different, but closely spaced, specimens. The comparability of these serial samples depends on the time interval between sampling and various in vivo and ex vivo factors, e.g. recent medications and specimen glycolysis, respectively. We present an approach for determining this maximum interval and demonstrate highly discrepant findings in two popular BGA systems.
Method:
Data repositories provided 5 years of ICU BGA glucoses and contemporaneously drawn CL glucoses from a Calgary ICU, equipped with IL GEM 4000 and CL Roche Cobas 8000-C702, and an Edmonton ICU, equipped with Radiometer ABL 800 and CL Beckman-Coulter DxC. To determine the maximum time interval that glucose differences would not be detected, we determined the mean absolute deviations [MAD] between the POC and CL glucoses for increasing sampling intervals: MAD = Σ |BGA glucose – CL glucose|/N
Result:
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
Department of Laboratory Medicine and Pathology, University of Alberta
Edmonton, Alberta, Canada
Objective:
For over the last 2 decades, our lab has been transforming the laboratory data accumulated by the US National Health and Nutrition Exam Survey (NHANES) into normal range graphs (www.mylaboratoryquality.com), sometimes stratified by waist circumference (WC), one of the most robust indicators of obesity. As the relationships between the NHANES tests and WC are profound, we used machine learning to provide salient clinical and diagnostic perspectives.
Method:
We used the regression trees machine learning algorithm to study the relationship of WC with the chemistry and hematology analytes of subjects enrolled in NHANES from 1999 to 2014. Glucose was forced to be the first split with its cut point chosen by the algorithm to give the smallest p-value when comparing the average of the WC values between the left and the right node.
Result:
The NHANES data set contains 6,812 male and 7,357 female Mexican Americans, 5,615 male and 5,828 female non-Hispanic Blacks, and 10,090 male and 10,368 female non-Hispanic Whites. The well recognized obesity-related tests, glucose and triglycerides (males and females) and ALT (males, not females) were classified as important as were GGT (increased), uric acid (increased) and albumin (decreased). To our satisfaction, the over-ordered, highly non-specific AST, was judged not to be WC-related. Many red blood cell associated tests (RBC count, iron, RDW, MCV, MCH) were related to WC, presumably through the action of the increased hepcidin produced by the liver.
Conclusion:
All of these WC-associated tests belong to a superset of tests and signs that includes the metabolic syndrome. Eventually, machine learning will enable logical diagnosis and therapy for the spectrum of obesity-related disease rather than its individual components.
Development of a Low-Cost Hemoglobin A1c Test for the Point-of-Care
Rice University Department of Bioengineering Houston, TX, USA
Objective:
Diabetes mellitus is a growing health burden in the United States that affects over 30 million Americans and costs more than $322 billion in health care expenditure annually. There is a need for an accurate and low-cost test to measure Hemoglobin A1c (A1c) that can be used at the point-of-care for diabetes management. Towards this aim, a fluorescent assay to measure A1c is being developed and optimized, with the goal of multiplexing with absorbance measurements to produce a quantitative reading of %A1c in a low-cost optical reader.
Method:
Four aptamer sequences specific to A1c were modified to form a tripartite structure-switching fluorescent aptamer complex through the addition of a 15 base-pair tail, a complementary fluorophore, and four complementary quenchers of various lengths, forming a total of sixteen aptamer-quencher pairs for screening. Initial screening was performed on half of the aptamer-quencher pairs by testing against a range of concentrations of A1c in solution (50-250μM), corresponding to 5-10%A1c given normal hemoglobin values, and measuring fluorescent response.
Result:
Two aptamer-quencher combinations show a quantitative, concentration-dependent fluorescent response to A1c relative to a no target control with correlation coefficients of -0.819 and -0.969 respectively, both of which are significant with p<0.001.
Conclusion:
This work presents a novel strategy to detect and quantify A1c from a low-cost fluorescent assay utilizing tripartite structure-switching aptamer complexes. Initial screening data demonstrates proof-of-concept utility of this approach, but further screening needs to be performed on the remaining aptamer-quencher pairs before selecting the best combinations to optimize and validate.
Induced Method by Dynamic Changes of Metabolism for Noninvasive Glucose Measurement
PhiScience Development UG, Schwerte, NRW, Germany
Objective:
The discomfort of blood sampling for conventional glucose measurements still remains a challenge. Our previous study, the Metabolic Heat Conformation (MHC) method, has been extended by the Induced Glucose Measurement (IGM) method which accounts for lipid metabolism and water content in body tissue. The device has been downsized to be user-friendly and electrocardiogram- and bio-impedance sensors have been added to existing thermal and optical sensors. The purpose of this study is to assess the practical usability of this supplemental method.
Method;
The IGM has been verified by two test protocols; 75g Oral Glucose Tolerance Test (OGTT) with 2 volunteers (1 time before glucose intake and 6 times every 30 minutes after glucose intake) and a clinical test with 12 volunteers (6 times a day in 2 days per each volunteer) were performed. A total of 144 duplicate measurements were obtained. Glucose results for IGM and SMBG were compared by applying Parkes Error Grid and Surveillance Error Grid. The IGM-model includes the ratio of carbohydrates and lipids in the energy balance and the metabolic rate can be deduced from pulse wave velocity, heat, and tissue parameters obtained from the integrated sensors.
Result:
The correlation coefficients of the regression(r) were 0.92 for the OGTT and 0.89 for the clinical test respectively at the confidential interval of 95%. The compliance pairs of ISO range was 78% for the clinical data.
Conclusion:
The IGM should not be directly compared to a conventional invasive method because the two methods have completely different analytical backgrounds and processing (sampling, calibration, etc.). The acceptance criteria for IGM should be further discussed in consideration of the user’s quality of life and convenience based on clinical effectiveness.
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
Diablo Clinical Research Walnut Creek, CA, USA
Objective:
A ready-to-use stable liquid Glucagon Rescue Pen (GRP; Xeris Pharmaceuticals, Inc.) auto-injector was compared to the GlucaGen® Hypokit® (GHK; Novo Nordisk) for the rescue treatment of insulin-induced severe hypoglycemia.
Method:
A Phase 3 non-inferiority, global, randomized, controlled, single-blind, 2-way crossover trial enrolled 132 adults with type 1 diabetes, ages 18 to 75 years, to compare subcutaneous 1 mg doses of the GRP versus GHK for the treatment of insulin-induced severe hypoglycemia in adults.
Result:
All evaluable subjects (100%) achieved a positive plasma glucose (PG) response (PG concentration >70 mg/dL (>3.88 mmol/L) or an increase in PG concentration > 20 mg/dL (>1.11 mmol/L) within 30 minutes of study drug injection) in both groups. Time to resolution of hypoglycemic symptoms and time to resolution of the overall feeling of hypoglycemia were comparable for both GRP and GHK. All subjects returned to euglycemia without the need for a second dose or alternative recovery measure. The most commonly reported adverse events (AEs) were: nausea, vomiting, and headache. The overall incidence of AEs was comparable in both groups. AEs were mild to moderate in severity and self-limited, and no serious adverse events (SAEs) occurred.
Conclusion:
Prompt treatment is critical in the rescue of severe hypoglycemic emergencies. GRP treatment resulted in successful PG recovery in all evaluable subjects, meeting the criteria for non-inferiority versus GHK. Additionally, GRP had a comparable time to resolution of hypoglycemic symptoms. Overall, no safety or tolerability concerns were noted. These results demonstrate that ready-to-use GRP is a viable alternative to currently available GHK.
It’s Time for Insulin and Pump-derived Measure of Engagement: Meal-Time Insulin BOLUS Score Outperforms the Self-Care Inventory
Kansas City University of Medicine and Biosciences Kansas City, MO, USA
Objective:
To compare the degree of association with concurrent hemoglobin A1c (A1c) for both the mealtime insulin BOLUS score (BOLUS) and the Self-Care Inventory-Revised (SCI-R). To further validate the BOLUS as an objective measure of engagement with self-management.
Method:
All individuals with contemporaneous 2-week insulin pump records, SCI-R, and A1c levels were included (n=105). Data were collected from pump uploads, electronic clinic intake forms, and electronic health records. To derive BOLUS, youths’ pump records were scored “1” if there was any carbohydrate-associated insulin bolus during standard mealtimes: Breakfast=0600-1000, Lunch=1100-1500, and Dinner=1600-2200; thus, the daily range of BOLUS is 0 to 3. Bivariate correlations and multivariable regressions were performed to measure associations for BOLUS and SCI-R, respectively, to A1c.
Results:
Among 105 participants, 56% were male, 92% non-Hispanic white, median age at diagnosis was 6.69 [IQR 4.0,9.1] years and median age for the sample was 15.54 [13.8,16.5] years. Median A1c was 8.5% [7.7%,9.6%]. Youths’ BOLUS and SCI-R were independently correlated with youths’ A1c level (r=-0.5796, p<0.001, and r=-0.319, p<0.001, respectively). In multiple regression analyses, youths’ BOLUS (b=-1.7, p<0.001) related more strongly to A1c levels than SCI-R (b=-0.006, p=0.353). In stepwise multivariable regression, the addition of SCI-R to a model containing BOLUS provided no additional explanatory power.
Conclusion:
The present work provides additional evidence of validity for the BOLUS as a measure of engagement with self-management. Reporting the BOLUS as an outcome metric in insulin pump data reports should be considered.
Sleep Duration Predicts Alteration in Metabolic Risk Factors
Department of Health Science and Technology, Aalborg University Aalborg, Nordjylland, Denmark
Objective:
A growing amount of evidence, from observational studies of sleep restriction and disorders, indicates that sleep duration is a central factor in the etiology of metabolic disturbances. These effects are important in the design of glucose monitoring systems and prediction models. The present study examines sleep duration and its relation to metabolic risk factors in people with diabetes.
Method:
We used data from several years of NHANES (2005–2013) and divided people into three groups based sleep duration. A GLM model (general linear model) was used to assess the association of sleep duration on metabolic factors in an adjusted analysis. We divided people into three groups based on previous reporting’s and self-reported sleep duration, short sleepers (<7 hours of sleep, n= 1,110), 7 hours of sleep (n= 552), and long sleepers (>7 hours of sleep, n= 976). A GLM model (general linear model) was used to assess the association of sleep duration on metabolic factors in an adjusted analysis. The metabolic factors were used as a dependent variable. In the adjusted analysis, the association was adjusted for BMI, age, gender, and ethnicity.
Result:
Glucose control was associated with sleep duration: >7 hours of sleep was found to be associated with an increased A1c level (est: 0.24 %-point, SE: 0.09, p<0.01) and fasting glucose concentration (est: 12.5 mg/dL, SE: 4.4, p<0.01).
Conclusion:
People sleeping more than 7 hours appear to have decreased glucose control. This knowledge could be considered when designing glucose monitoring systems and prediction models.
Development of a Gestational Diabetes Self-Management and Remote Monitoring Mobile Platform
University of Cape Town
Cape Town, Western Province, South Africa
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
South-Westphalia University of Applied Sciences, Interdisciplinary Centre for Life Sciences
Iserlohn, NRW, Germany
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
BD Technologies Research Triangle Park, NC, USA
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”
Rensselaer Polytechnic Institute Troy, NY, USA
Objective:
The Do-it-Yourself (DIY) “Loop” community, spurred on by the slow progress in commercializing automated insulin dosing devices, has gained popularity among individuals who want to achieve greater control over their blood glucose. To date, however, there has been no analysis of the loop algorithm in the context of feedback control theory.
Method:
We performed a detailed analysis of the Loop algorithm, which consists largely of four contributions:
• Insulin effect on glucose
• Carbohydrate effect on glucose, modelled as a ramp
• Retrospective correction effect of glucose predictions to account for un-modeled effects
• Momentum correction effect, to account for recent blood glucose trends
Result:
Loop is in a class of model predictive control (MPC) algorithms, where a model is used to forecast the effect of current and future insulin actions on future glucose values. Classical model predictive control solves this as an optimization problem over a prediction horizon, with “decision variables” of the insulin infusion rate changes over a shorter control horizon. The Loop algorithm is similar to “coincidence point” MPC, since a single value of glucose is considered at the end of the prediction horizon. Because only one insulin infusion rate is calculated (spread over a 30 min interval), the Loop algorithm does not require solution of an optimization problem. The correction for the mismatch between glucose predictions and measurements at each time step is similar to that often used in MPC. The primary safety check is to confirm that the glucose prediction does not go below a hypoglycemic threshold over the prediction horizon. Additionally, conservative estimation of carbohydrate absorption rate avoids excessively high glucose predictions.
Conclusion:
The Loop algorithm was examined and found to be similar to coincidence point MPC. A control “system” is much more than a specific control algorithm, and includes additional components. The Loop algorithm would benefit from additional safety and fault detection components, to ensure validity of the blood glucose measurements fed to the algorithm, for example.
Assessment of a “DIY Looping” Algorithm using the UVa/Padova Metabolic Simulator
Rensselaer Polytechnic Institute Troy, NY, USA
Objective:
Due to few available closed-loop artificial pancreas (AP) systems, “do it yourself (DIY) looping” has become a popular means of achieving closed-loop glucose control. The objective of this work is to compare the Loop algorithm to existing, well-studied algorithms. A comparative simulation study was conducted using the UVa/Padova (UVA) Metabolic Simulator. Our implementation of a popular DIY looping (DIY-Loop) algorithm was compared against a proportional–integral–derivative (PID) feedback controller using published values, with and without a meal bolus provided. A basal-bolus regimen (BB) was also considered.
Method:
Three scenarios were investigated: (i) overnight control (11pm-7am), (ii) control after a single 9:00am meal, and (iii) 36-hours of continuous control with six total meals. Adult and adolescent patients 001-010 were investigated with average time-in-range (70-180 mg/dL) as the performance metric.
Result:
In the overnight scenario, all methods investigated achieved 100% of time-in-range. In the meal scenario with a meal bolus provided, the average percent time-in-range was 64/70/72% for BB, PID, and DIY-Loop, respectively. When no meal bolus was provided, the average time-in-range was 60% for both PID and DIY-Loop. In the 36-hour scenario, with meal bolus provided, average time-in-range was 66/69/75% for BB, PID, and DIY-Loop, respectively. Time spent below 50 mg/dL in this scenario was 0/115/33 min for BB, PID, and DIY-Loop, respectively. In the 36-hour scenario, with no meal bolus, time in range was 56/61% for the PID and DIY-Loop, respectively.
Conclusion:
A DIY looping control algorithm was compared with basal-bolus and PID using the UVA simulator and three scenarios of varying complexity. Comparable performance between all algorithms was seen in the overnight and single meal, with no meal bolus, scenarios. In the 36-hour scenario, the looping algorithm achieved the longest time- in-range, both with and without a meal bolus. The results of this simulation study show that, in most cases, a DIY looping algorithm achieves the most time within range, relative to a basal-bolus regimen and PID control.
Prediction of Postprandial Glycemic Response from Meal Models of Varying Complexity
Rensselaer Polytechnic Institute Troy, NY, USA
Objective:
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
Center for Health Information and Decision Systems, Robert H. Smith School of Business
University of Maryland, College Park College Park, MD, USA
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
Modelling of Glucose Dynamics and Estimation of Insulin Sensitivity from Glucose Data Only
School of Engineering, University of Warwick Coventry, West Midlands, United Kingdom
Objective:
The measurement of insulin is laborious, expensive, and subject to significant inter-assay variability. Nevertheless, the vast majority of mathematical models for postprandial glucose metabolism and the estimation of insulin sensitivity rely upon the measurement of insulin, hindering their use in clinical practice. We, therefore, developed a mathematical model for the description of postprandial glucose dynamics and estimation of insulin sensitivity utilizing only glucose data.
Method:
A total of 66 postprandial glucose and insulin responses from healthy, young, non-obese subjects consuming three identical mixed meals in a single day were utilized to identify the well-established oral minimal model (OMM), yielding a description of glucose dynamics and estimate of insulin sensitivity from all responses. A newly developed adaptation of the OMM was subsequently identified from the glucose responses only. It utilizes a novel description for the meal related appearance of glucose and introduces a description for glucose dependent stimulation of the “active” insulin effect.
Result:
The model using glucose data only yields a similar description of glucose dynamics in comparison to the OMM, as assessed by the root mean-squared error (mean 0.28 vs 0.29 mmol/L, p>0.5). The newly introduced parameter governing the glucose dependent stimulation of the insulin effect in the glucose only model correlates well with OMM estimated insulin sensitivity (r = 0.61, p < 10-6).
Conclusion:
This work is the first attempt at estimating insulin sensitivity from glucose data only and shows promising results. It uses minimal data collected under realistic experimental conditions. We therefore argue that this approach has the potential for the widespread quantification of insulin resistance in people with pre- and type 2 diabetes in clinical practice.
Hospital and Outpatient Insulin Management with a use of Digital Therapeutics
Voluntis, Medical Affairs
Cambridge, MA, USA
Objective:
Review inpatient and outpatient hyperglycemia, hypoglycemia, and glycemic management and how it applies to eGlycemic Management Systems (eGMS) and FDA cleared self-titration apps.
Method:
Present a case study with outcomes relating to use of eGMS such as EndoTool and FDA cleared self-titration app such as Insulia.
Result:
The inpatient eGMS such as EndoTool helps patients achieve glycemic control and optimize patient outcomes. Data show 90% reduction in hypoglycemia readings with the future insulin activity factored into the algorithm. The days between readings represents more than 11 statistical years without a hypoglycemic event. Use of outpatient FDA cleared self-titration app, such as Insulia, contributes to patient engagement.
Conclusion:
The inpatient eGMS such as EndoTool helps patients achieve glycemic control and optimize patient outcomes. Upon hospital discharge, effective self-titration of insulin can be done with the use of a FDA cleared app such as Insulia.
Outcomes in Pump- and CGM-Naïve Subgroups in the International Diabetes Closed-Loop (iDCL) Trial
Department of Pediatrics, Division of Pediatric Endocrinology and Diabetes, Stanford University School of Medicine
Stanford, CA, USA
Objective:
We investigated potential benefits of automated insulin delivery in a technologically naïve sub-population with type 1 diabetes (T1D) using multiple daily injections (MDI) or insulin pumps without a continuous glucose monitoring system (CGM).
Method:
In a 6-month randomized, multicenter trial (RCT), 168 individuals were assigned to closed loop control (CLC, Control-IQ, Tandem Diabetes Care) or sensor-augmented pump (SAP) therapy. The trial included a 2-8 week run-in phase (depending on baseline device use) to train participants on study devices. We compared glycemic outcomes in the RCT among four subgroups: baseline insulin pump and CGM users (pump+CGM, 61%), pump-only users (18%), MDI and CGM users (MDI+CGM, 10%), and MDI users without CGM (MDI-only, 11%).
Result:
Among participants in the CLC group, CGM time in range (TIR), 70-180 mg/dL, improved regardless of past pump and/or CGM use from 61±17% to 71±12%, and TIR in the SAP group remained unchanged at 59±14%. Mean TIR improved from baseline in the four subgroups using CLC: pump+CGM, 62% to 73%; pump-only, 61% to 70%; MDI+CGM, 54% to 68%; and MDI-only, 61% to 69%. The reduction in time below 70 mg/dL (TB70) from baseline was comparable among the four subgroups: pump+CGM, 53 to 22 minutes; pump-only, 67 to 31 minutes; MDI+CGM, 22 to 13 minutes; and MDI-only, 41 to 19 minutes. There was no interaction effect with baseline device use for TIR (P=0.67) or TB70 (P=0.77).
Conclusion:
There was a consistent benefit in patients with T1D when using CLC, regardless of baseline insulin delivery modality or CGM use. These data suggest that this CLC system can be considered across a wide range of patients.
Clustering and Stratification of CGM Daily Profiles in the International Diabetes Closed-Loop (iDCL) Trial
University of Virginia Center for Diabetes Technology Charlottesville, VA, USA
Objective:
Introduce a novel feature learning approach to the analysis of continuous glucose monitor (CGM) daily profiles in the iDCL trial applicable to treatment optimization and decision-making support.
Method:
Protocols iDCL-1 (NCT02985866) and iDCL-3 (NCT03563313) had N5/N8 participants randomized 1:1/2:1 to Closed Loop Control (CLC) or Sensor Augmented Pump (SAP) for 3/6-months. The iDCL-1 data were used to develop a novel feature learning technique extracting low-dimensional temporal features from daily CGM profiles and clustered using K-means. With the trained feature extractor and clusters, the daily CGM profiles in iDCL-3 were classified. The utility of the methodology was verified by analysis of cluster association with established glycemia metrics for all subjects and by contrasting the control and experimental groups.
Result:
A 4-cluster classifier was trained on iDCL-1 data and applied to iDCL-3 daily CGM profiles. Cluster-specific glycemia metrics were similar between iDCL-1 and iDCL-3. The clusters are characterized as (data shown: iDCL-1/iDCL-3 mean±SD glucose in mg/dL): (Cluster1) tight/intense control during 3am-noon (Period1) (135±27/134±22) and noon-3am (Period2) (137±17/138±15); (Cluster2) tight Period1 control (148±30/147±27) with moderate Period2 hyperglycemia (183±19/182±19); (Cluster3) moderate Period1 hyperglycemia (181±41/175±38) with moderate Period2 hyperglycemia (180±21/179±21); and (Cluster4) Period1 hyperglycemia (205±53/198±51) and even higher Period2 hyperglycemia (251±30/250±29). The new classifier is sensitive and differentiates subject groups by treatment. For example, CLC subjects had less hyperglycemic Cluster4 daily profiles than SAP subjects in iDCL-1/iDCL-3 (3.95/7.95 % for CLC vs. 8.61/13.28 % for SAP; p<0.05 for CLC vs. SAP).
Conclusion:
A novel feature-learning technique for temporal feature extraction trained on iDCL-1 and applied to iDCL-3 data classifies daily CGM profiles into four separable clusters which differentiate CGM profiles between different times of the day, 3am to noon vs. noon to 3am.
Effect of Dawn Phenomenon on Glucose Infusion Rate during Glucose Clamp Studies In Subjects with Type 1 Diabetes
ProSciento, Inc. Chula Vista, CA, USA
Objective:
Dawn phenomenon is a well-recognized early morning increase in blood glucose with its greatest effect generally recognized between 4-8am in diabetic patients. This physiologic phenomenon could theoretically impact the findings, particularly around the end of action, in glucose clamp studies evaluating pharmacodynamic (PD) effects of new investigational products (IP) with decreased glucose requirements to maintain a target blood glucose level. We measured changes in glucose infusion rate (GIR) requirements during glucose clamp studies with long-acting insulin to assess this effect.
Method:
Differences between overnight (12–4am), early morning (4–8am), late morning (8am–12pm), and afternoon (12-4pm) mean GIR requirements were compared in 21 euglycemic clamps with type 1 diabetic (T1D) subjects receiving the same long-acting insulin. We selected 36-and 42-hour clamps in which the study insulin was administered in the morning, was known or expected to have a flat time-action profile, and had not reached end of action defined as a consistent increase in blood glucose without GIR.
Results:
Our study population included T1D adult (mean age 37.1±11.2) subjects diagnosed for >12 months. The overnight mean GIR (mg/kg/min) was 0.90±0.52 (mean±SD), early morning mean GIR was 0.64±0.50, late morning mean GIR was 0.66±0.47, and afternoon mean GIR was 1.00±0.41. Comparisons of mean GIR were statistically significant between overnight and early morning (p=0.006), early morning and afternoon (p<0.001), and late morning and afternoon (p=0.001).
Conclusion:
Controlled glucose clamp studies provide confirmation of decreased GIR requirements during early morning and into late morning, with increases in GIR later, likely related to dawn and extended dawn phenomenon in T1D subjects. This effect should be anticipated in evaluating PD effects of new investigational products.
Insulin Pump Accuracy Evaluation – an Update for Bolus Delivery
Institute for Diabetes Technology, Research and Development mbH at the University of Ulm
Ulm, Germany
Objective:
Insulin pump accuracy is of current interest especially in view of closed loop systems. Accuracy results were already presented for different insulin pumps; this study is an update adding recently introduced pumps.
Method:
Three durable insulin pumps were tested in an experiment based on IEC 60601-2-24: MiniMed® 670G, mylifeTM YpsoPump® and t:slim X2. Bolus accuracy was evaluated for boluses of 10 U, 1 U and 0.1 U; in addition, the delivery speed of a 10 U bolus was assessed. Pumps were installed next to a balance and insulin was delivered through the infusion set into a water-filled beaker placed inside the balance chamber. Delivery was calculated from weight gain and compared to target boluses. Per model 9 data sets (3 pumps in 3 repetitions) were obtained; one data set comprised 25 (0.1 U and 1 U) or 12 boluses (10 U). As there are no acceptance criteria in IEC 60601-2-24, ±15% from target was considered relevant.
Result:
Total bias of 10 U boluses ranged from -0.8% to 0.7% and, for all pumps, all individual boluses were within ±15% from target. For 1 U boluses, total bias ranged from 0.3% to 1.9% with 100% of boluses within ±15%. Delivery of the bolus of 0.1 U showed total deviations of -2.3% to 6.0% and 85% to 96% of boluses did deviate less than 15% from the intended bolus. Delivery of the 10 U bolus took 19 seconds to 6:42 minutes.
Conclusion:
With these results, the new pumps fit well into the picture of previously tested durable insulin pumps. In general, accuracy is lower with smaller boluses whereas for boluses of 1 U and larger delivery seems reliable.
Post-market Surveillance of the Accuracy of 18 Blood Glucose Monitoring Systems Available in Europe Based on ISO 15197:2013
Institute for Diabetes Technology, Research and Development mbH at the University of Ulm
Ulm, Germany
Objective:
In this study, system accuracy of 18 current-generation SMBG systems from different manufacturers was evaluated based on ISO 15197:2013 with 1 test strip lot each. Manufacturers were selected by their respective market share in Europe.
Method:
The following systems were tested with one test strip lot each: ABRA (A), Accu-Chek® Guide (B), AURUM (C), CareSens™ Dual (D), CERA-CHEK™ 1Code (E), Contour® next ONE (F), eBsensor (G), FreeStyle Freedom Lite (H), GL50 evo (I), GlucoCheck GOLD (J), GlucoMen® areo 2K (K), GluNEO® (L), MyStar DoseCoach® (M), OneTouch® Verio Flex (N), Pic GlucoTest (O), Rightest™ GM700S (P), TRUEyou (Q), and WaveSense JAZZ™ Wireless (R). Each system was tested using capillary blood samples from 100 different subjects. Compliance with accuracy criterion A of ISO 15197 (≥95% of results within ±15% or ±15mg/dl of the comparison method’s results for BG concentrations above or below 100 mg/dl, respectively) was assessed. Duplicate measurements were performed with the BGMS. Glucose oxidase-based YSI 2300 STAT Plus and hexokinase-based Cobas Integra® 400 plus were used as comparison methods. Additionally, mean absolute relative differences (MARD) were calculated.
Result:
In total, 14 of 18 systems had ≥95% of results within ±15% or ±15mg/dl of the respective system manufacturer’s comparison method’s result. Individual systems showed 89% to 100% of results within these limits. Smallest deviation limits within which ≥95% of results were found ranged from ±7.7% or ±7.7mg/dl to ±19.7% or ±19.7mg/dl. MARD ranged from 3.41% to 9.79%.
Conclusion:
Although only current-generation systems available on the market were used, only 78% of systems met the minimum system accuracy criterion A of ISO 15197 with the tested test strip lot. Considerable differences in accuracy were found.
Light Control of Insulin Delivery: High Insulin Density 2nd Generation Materials
Division of Pharmacology and Pharmaceutical Sciences, University of Missouri-Kansas City
Kansas City, MO, USA
Objective:
Our objective is to control the delivery of insulin in a continuously variable manner, without the use of a pump or any physical connection between the outside and the inside of the patient.
Method:
To accomplish this objective, we are developing the PhotoActivated Depot (PAD) approach, in which insulin containing materials are injected into the skin, where they remain, inactive, until stimulated through the skin with an LED light source to release insulin. We have previously shown that this approach can work, with insulin being released in-vivo from an injected depot using pulses of light from an external LED light source, followed by blood glucose reduction.
Result:
In the currently reported work, we have designed, synthesized, and tested 2nd generation materials that significantly improve required properties. These properties include insulin density, which we have increased ten-fold. This allows for longer depot duration and greater ease of insulin release. In addition, we have incorporated light cleaved links with longer wavelengths, allowing for greater tissue penetration.
Conclusion:
With our new materials, we are nearing the performance levels required for human subjects. Key parameters include the amount of insulin contained within an injection volume and the amount of insulin released with a tolerable dose of light.
Glucose Monitoring – What? How? Why?
Iridium Consultancy & Technologies Ltd.
Herzliya, Israel
Objective:
Trillions of glucose measurements have been conducted by hundreds of millions of people since the introduction of the first blood glucose meter about 38 years ago. Despite that, basic questions such as “how often”, “when”, “what to do with the numbers”, “which device” still remained vague and unclear. Glucose accuracy still considered a major issue yet its meaning is persistently blurred.
Method:
A thorough research about the history of the development, comparisons between different glucose monitoring devices through the years, and focus on potential causes and reasons for discrepancy between readings was performed and analyzed. Parameters like technology, frequency of measurements, human factors, assessing the numbers and behavior based on the data, were taken in consideration. The term “accuracy” was also analyzed, among others with regard to its impact on understanding the measurements’ outcome. Methods for evaluation of the accuracy, clinical, and statistical tools were also studied.
Result:
Different ways of conducting clinical trials cause altered results. Misunderstandings regarding the nature of measurement (ex. comparing plasma blood vs. interstitial fluid) also led to inconsistent results. Human factors (poor maintenance of disposables, way of measurement, etc.) impacts results. Discrepancies of 12-30%, and some extremes of 40-80%(!), were observed between devices when the measurements were conducted by professionals. Testing the 1st drop from an unwashed finger led to catastrophic errors.
Conclusion:
Manufacturers should focus more on reducing the impact of human factors rather than “fighting” to achieve 1% more accuracy in the device itself. Better evaluation of
Model Based Assessment of Once Daily iGlarLixi Administration Timing on Glucose Control in Subjects with Type 2 Diabetes: An In-Silico Study
Center for Diabetes Technology, UVA Charlottesville, VA, USA
Objective:
iGlarLixi is a fixed ratio combination of insulin glargine and glucagon-like peptide 1 receptor agonist lixisenatide. Administered once daily, it was shown to be safe and efficacious for type 2 diabetes (T2D) control. We proposed to compare the glycemic control achieved by pre-breakfast versus pre-dinner administration of iGlarLixi in a newly built insulin treated T2D in-silico population.
Method:
iGlarLixi pharmacokinetics (PK) and pharmacodynamics (PD) were modelled and parametrized using both intravenous and oral glucose challenges with or without lixisenatide in T2D, obtaining model structure and initial parameter estimates, and fitting model predictions to observed glucose, insulin, c-peptide and glucagon responses. These PK-PD models were integrated in a pre-existing T2D simulation platform. Further data (14-days of CGM pre/post a 24 week daily lixisenatide) refined the augmented simulation platform parameter distribution (in-silico
Result:
Pre-breakfast versus pre-dinner administration of iGlarLixi resulted in 4.19±0.12 (mean ± standard-error) decrease of HBGI in the morning but 2.09±0.07 increase in the evening and a decrease of LBGI in the morning and evening of 0.15±0.014 and 0.27±0.03, respectively. No major differences were observed beyond these time periods.
Conclusion:
Early simulation results using our platform indicate that pre-breakfast administration of iGlarlixi improves hyperglycemia protection in the morning relative to pre-dinner administration, alongside a slightly decreased overall risk of hypoglycemia. Pre-dinner administration improves evening protection against hyperglycemia.
Effectiveness of Social Media to Disseminate Education about Foot Ulcers among Patients with Type 2 Diabetes Mellitus
University of Maryland Baltimore Washington Medical Center Glen Burnie, MD, USA
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
Texas A&M University, Department of Computer Science and Engineering
College Station, TX, USA
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.
A Comparison between Aerobic and Strength Training Exercise on the Reduction of Cardiovascular Disease
University of Texas Medical Branch Galveston, TX, USA
Objective:
We desired to know, in healthy patients ages 18 and older, does strength training or aerobic exercise confer more benefit for decreasing risk factors of CVD?
Method:
We performed a systematic review to analyze the results of 20 studies within the last 10 years, to determine whether aerobic or strength training conferred greater benefits in reducing CVD risk factors.
Result:
Individually, we found that aerobic exercise and resistance training reduce cardiovascular risk factors. Direct comparison of strength versus aerobic training yielded conflicting results. In our review, many papers compared a combination exercise program to these individual forms of exercise and the results of these studies unanimously showed major benefit using a combination exercise program and, in most cases, showed greater benefit than a solely strength or aerobic exercise program.
Conclusion:
Overall, while it is difficult to discern whether strength versus aerobic exercise confers more benefit, our research showed that any form of exercise confers some benefit in reducing CVD risk factors and health care professionals should stress the importance of regular exercise to patients in order to decrease the epidemic of CVD.
Impact of Insulin Stability on Intraperitoneal Insulin Catheter Obstruction: A Comprehensive Analysis in Rodents and Type 1 Diabetic Patients
Department of Pharmaceutical Sciences, University of Connecticut
Storrs, CT, USA
Objective:
Continuous intraperitoneal insulin infusion (CIPII) with an implantable pump shows a rapid profile of insulin adsorption as well as clearance, improving glucose control. However, intraperitoneal catheter obstruction remains an occasional issue and hinders the widespread acceptance of this technology. It has been hypothesized that catheter obstructions could be promoted by the formation of insulin amyloid fibrils at the catheter tip, followed by inflammatory cell deposition. By obtaining a rigorous understanding of insulin stability in catheters, it may be possible to develop mitigations and reduce the overall catheter complication rate with CIPII.
Method:
The remaining insulin was collected from patients’ pumps and compared with fresh insulin, in the aspects of molecular size and structure. Excised catheter blockages from patients were analyzed histologically. A rodent model of intraperitoneal catheter obstruction was developed and the extent of catheter obstruction was compared, in a time-dependent manner, both macroscopically and microscopically between the fresh insulin groups and control groups (i.e., only administered diluent).
Result:
There was a slight change in insulin size and structure between fresh insulin and the samples collected from the patients’ pumps. In addition, amyloid proteins were observed in some of the tip deposits explanted from patients. In the rodent model, both the fresh insulin groups and control groups showed similar development and progress of catheter obstruction. No insulin precipitation was observed in the catheter lumen area in the rodent model.
Conclusion:
Insulin infusion by itself is unlikely to be the leading cause of catheter obstructions since they also occurred in the control groups where only diluent was administered. Nevertheless, transformed insulin after staying in pump reservoir could contribute to enhance inflammatory reactions at the catheter tip.
Size and Tip Configuration-dependent Foreign Body Reaction to Intraperitoneal Insulin Catheters
Department of Pharmaceutical Sciences, University of Connecticut
Storrs, CT, USA
Objective:
Intraperitoneal catheters are used in continuous intraperitoneal insulin infusion (CIPII) to directly deliver insulin into the intraperitoneal space. Despite the many clinical advantages of CIPII, the incidence of catheter obstructions (including lumen occlusion, catheter tip blockages, catheter tip encapsulations, and full catheter encapsulations) has decreased the acceptance of this technology. It is hypothesized that through modification of the catheter physical properties, such as size and configuration, it may be possible to mitigate the foreign body reaction (FBR) and reduce the obstruction rates. This may also provide new insight into the design of implantable devices in a variety of other applications.
Method:
Catheters with either different outer diameters (0.1 inch vs. 0.048 inch) or different tip configurations (cone-shape vs round-shape) were implanted intraperitoneally into rats and then explanted at different times. A gross macroscopic analysis was performed to determine the location as well as the extent of obstructions, followed by various histological staining of the blockages and encapsulation tissues. Lumen occlusion was identified using polarized light microscopy.
Result:
Although round-tip catheters showed less inflammation, lumen occlusions occurred, whereas cone-tip catheters showed more inflammation but no lumen occlusion (this may be due to the special design of the cone-shaped tip significantly reducing the back pressure in the lumen). The small-sized catheters triggered less FBR, however they required fewer cells to form encapsulations, and thus the extent of obstructions for both big- and small- sized catheters were similar.
Conclusion:
Tip configuration and size influences the FBR to catheters. However, catheter obstructions are not dependent on a single factor. A good catheter design requires reducing the FBR as well as maintaining functionality to achieve long-term performance.
Investigation of Insulin Formulations by Analytical Methods – Overview and Possibilities for Quality Control
South-Westphalia University of Applied Sciences, Interdisciplinary Centre for Life Sciences
Iserlohn, Germany
Objective:
Human insulin and its analogues play an important role in the regulation of carbohydrate metabolism for people with diabetes. For insulin in blood plasma, chromatographic assays with UV detection or mass spectrometry have shown high sensitivity and the potential of separation and detection of insulin and its analogues. While endogenous insulin exists as monomers, these assemble into dimers, which further can form hexamers coordinated by zinc ions, as found in therapeutic formulations.
Method:
Besides the established methods, a variety of analytical assays for total protein analysis by using dyes with either colorimetry or fluorescence detection have been investigated for the quantification of insulins in commercial formulations. Different insulins and their formulation components such as phenolic substances have been studied by UV spectroscopy. Quantitative analysis is only possible in combination with chromatographic separation techniques. For isolating the insulins from the formulation, ultrafiltration has been suggested by us. For identification and quality control of both formulated and pure insulins, matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry and infrared-attenuated total reflection (ATR) spectroscopy have been successfully studied.
Result:
While for most assays, total protein determination is possible regardless of protein misfolding, identification is uniquely possible using MALDI-TOF mass spectrometry based on molecular masses independent from fibrillation. Concerning quality control, infrared spectroscopy allows the detection of changes in the secondary structure of the insulins within long-term studies under deviating storage conditions other than recommended.
Conclusion:
From all methods studied, infrared spectroscopy as a fast, low-cost, and reliable technique can provide information on the protein secondary-structure and changes negatively correlated to the insulin bioactivity. Related infrared bands were analyzed by applying various procedures such as band deconvolution, analysis of band shifts or score plots, after principal component analysis.
Evaluation of the Analytical Performance of YSI 2900D Compared to YSI 2300 STAT Plus™ for Glucose Concentration Measurements
ProSciento, Inc. Chula Vista, CA, USA
Objective:
The aim was to evaluate the analytical performance of the YSI (YSI, Inc. Yellow Springs, OH USA) 2900D (YSI 2900) laboratory instrument using the validated YSI 2300 STAT Plus™ (YSI 2300) medical device as the reference method.
Method:
Serum controls (YSI and NIST SRM 965b) and blood and plasma samples were used to assess the precision, accuracy (bias), and linearity of both analyzers for glucose concentration measurements. Human blood (4 subjects, 38.5±5.8 years of age, BMI of 33.8±2.1 kg/m2) was collected to prepare blood and plasma samples with low (1-70 mg/dL), mid (70-125 mg/dL) and high (125-900 mg/dL) glucose concentrations. Both analyzers employed a glucose oxidase amperometric biosensor and were configured with the same parameters. Daily membrane integrity and linearity checks were performed prior to conducting sample analyses. Data are mean±SD.
Result:
Both instruments passed the daily performance checks prior to sample analyses. The YSI 2900 analyzer demonstrated similar within-run precision as the YSI 2300 (CV<2% for each analyzer and each serum control tested). A total of 80 blood samples (glucose range: 3.7–766.3 mg/dL) and 80 plasma samples (glucose range: 4.7– 880.7 mg/dL) were analyzed in triplicate with both analyzers. Paired results from the YSI 2900 and YSI 2300 showed a high correlation coefficient (R2>0.99) and similar regression statistics for both blood and plasma samples. Both the YSI 2900 and the YSI 2300 analyzers met accuracy (bias) criteria and a Clarke error grid analysis demonstrated clinical accuracy with 100% of blood and 100% of plasma results in zone A.
Conclusion:
Results from this study demonstrates that the YSI 2900 analyzer is analytically comparable to the YSI 2300 for a wide range of blood and plasma glucose concentrations.
The Effect of Digital Intervention on Glycemic Control in Users with Diabetes
DarioHealth Corp Caesarea, Israel
Objective:
To examine whether digital intervention contributes to better diabetes management. The platform launched in the end of 2018 includes the use of DarioTM blood glucose meter with the mobile App, weekly progress reports, coaching webinars, feedback alerts, relevant content delivery, and follow-up with a dedicated professional via in-App chat, emails and phone calls.
Method:
A population of active users measuring with Dario for at least a month before enrolling to the Dario EngageTM platform (Baseline) and having at least 3 months data on the Dario Engage was evaluated. Clinical outcomes examined were: Blood glucose average (BGavg), % in-range measurements (70-180 mg/dL) per the total number of measurements, and the percentage of the population that reduced their blood glucose average below 140 mg/dL following 3 months of digital intervention.
Result:
A total of 162 users improved their percentage glucose in-range following 3 months by 6% (165±48.7 vs. 155±42.1). Almost half [45% (73 out of 162)] of users reduced their BGavg under 140mg/dL (equivalent to estimated A1C of 6.5%) following 3 months. Subgroup analyses of 101 users out of 162 started with BGavg >140mg/dL revealed an increase of 19% for in-range measurements of the total measurements when compared to the baseline (65% vs. 54%). Moreover, 51 users in high risk (started with BGavg >180 mg/dL) increased their percentage in-range measurements by 38% (48% vs. 35%) and BGavg by 14.3% (192±39 mg/dL vs. 224±38 mg/dL).
Conclusion:
Patients using a digital intervention platform have the potential to modify their behavior, enhance adherence to diabetes management, and demonstrate better glycemic control.
A Digital Lifestyle Program to Support Outpatient Treatment of Type 2 Diabetes: A Randomized Controlled Trial
Akureyri Hospital, University of Akureyri Akureyri, Iceland
Objective:
Lifestyle is important in the treatment of type 2 diabetes. The aim of this study was to investigate whether a lifestyle program through a smartphone application could affect treatment outcomes at an endocrinology outpatient clinic.
Method:
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
Illinois Institute of Technology, Department of Biomedical Engineering
Chicago, IL, USA
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
University of Connecticut, Department of Pharmaceutical Sciences
Storrs, CT, USA
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
University of Virginia Charlottesville, VA, USA
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
Research Division, Joslin Diabetes Center and Department of Pediatrics, Harvard Medical School
Boston, MA, USA
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.
Predictors of Post-Prandial Hypoglycemia in Type 1 Diabetes and Continuous Subcutaneous Insulin Infusion
Steno Diabetes Center North Denmark, Aalborg University Hospital; Department of Health Science and Technology, Aalborg University Aalborg, Denmark
Objective:
New ultra-fast-acting insulins show promising results in reducing post-prandial glucose but, unfortunately, they increase the risk of hypoglycemia. This study sought to investigate predictors of post-prandial hypoglycemia.
Method:
A logistic regression model was used to analyze 24,185 meals from type 1 diabetes patients (n=427) enrolled in the Onset 5® trial by Novo Nordisk using continuous subcutaneous insulin infusion and continuous glucose monitoring (Dexcom G4). In an equal distribution of breakfast, lunch and main evening meals, the participants experienced 1,713 hypoglycemic episodes (interstitial glucose ≤ 54 mg/dL) within three hours after start of meal. Participants were on average 43 (SD:15) years old with an average diabetes duration of 24 (SD:12) years.
Result:
Results showed that primarily being young (OR: 1.29, 95% CI: 1.09-1-51), main evening meal (OR: 1.15, 95% CI: 1.01-1.30), rate of insulin change compared to previous meal of same type (OR: 1.28, 95% CI: 1.09-1.51), hypoglycemia at previous meal of same type (OR: 1.81, 95% CI: 1.55-2.11), drop in interstitial glucose before meal (OR: 1.51, 95% CI: 1.32-1.73) increased the risk of post-prandial hypoglycemia. On the other hand, A1c (OR: 0.80, 95% CI: 0.72-0.89), being an American (OR: 0.82, 95% CI: 0.72-0.92), and minimum interstitial glucose before meal (OR: 0.78, 95% CI: 0.77-0.80) decreased the risk of post-prandial glucose hypoglycemia.
Conclusion:
The most pronounced predictors of post-prandial hypoglycemia were previous hypoglycemia, drop in interstitial glucose, and rate of insulin change. Furthermore, baseline adjustments, such as, for age, A1c, and region are important. These findings aid to development of models for prediction of post-prandial hypoglycemia, which would be a tremendous help for the patient to guide on bolus dose or meal adjustments.
“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
The Jefferson Artificial Pancreas Center, Department of Anesthesiology, Thomas Jefferson University
Philadelphia, PA, USA
Objective:
The purpose of this study was to compare tissue response and insulin bolus spread using straight and angled insulin infusion sets (IIS).
Method:
IIS with different cannula insertion angles (Medtronic Silhouette, 30° and Quickset, 90°) were inserted subcutaneously every other day for 2 weeks in 11 swine and connected to an insulin pump (basal/bolus pattern). After 2 weeks, an insulin/x-ray contrast agent bolus was infused through the IIS while recording a pressure profile, evaluating maximum tubing pressure (pmax), and the area under the pressure curve (AUC). The tissue-cannula specimen was excised and imaged using micro-CT to measure insulin bolus surface area, volume, and surface-area-to volume ratio (SA:V). Specimens were processed for histopathological analysis of area of inflammation (AI) and thickness of inflammatory layer (LT) surrounding the cannula. Data were analyzed using ANOVA GLM, Kruskal Wallis, and post-hoc Bonferroni correction.
Result:
Insulin infusion through a straight cannula caused significantly higher pmax (p=0.005) and AUC (p=0.014). Bolus SA (30°: 314.0±84.4 mm2 vs 90°: 228.7±99.7 mm2, p<0.001) and V (30°: 198.7±66.9 mm3 vs 90°: 145.0±65.9 mm3, p=0.001) were larger using an angled infusion set with no effect of angle on SA:V. Bolus SA and V decreased significantly over infusion set wear time independent of insertion angle (p<0.05). Straight cannulas caused a greater mean AI (30°: 90.1±17.0 mm2 vs 90°: 70.6±21.0 mm2, p<0.001) and LT (30°: 67.2±24.3 vs 90°: 90.1±16.8 mm, p<0.001). As the LT increases, pmax increases and bolus volume decreases.
Conclusion:
Although a straight IIS is clinically preferred due to easier insertion, our data suggest that angled cannulas elicit less inflammation and deliver a bolus of insulin with lower tubing pressure and greater SA and V.
A Non-Invasive Glucose Measurement Device Based on Ultraviolet Light Scattering
Vital USA Inc.
West Palm Beach, Florida
Background:
Availability of non-invasive and cost effective monitoring of glucose levels will change the lives of millions of people with type 1 and type 2 diabetes worldwide. The newly developed VitalDetect device (Vital USA, Palm Springs, FL) measures 7 vital signs including non-invasive tissue glucose from the finger tissue. The device utilizes reflective light scattering at 395 nm in conjunction with additional traditional spectral analyses at frequencies of 660 nm and 940 nm, respectively, to calculate the non-invasive tissue glucose levels.
Method:
The technology employed by VitalDetect utilizes the pulse rate signal from the 940 nm emitter to filter and evaluate the ultraviolet signal at 395 nm. The signals obtained at 940 nm serves as internal reference for the 395 nm signal, a similar approach as using the 660 nm signal for assessment of oxygen saturation in pulse oximetry. The device is controlled through the Vital App and communicates all results via Bluetooth (BLE) to the smart phone. This technology is licensed from Sanmina (Huntsville, AL). The non-invasive glucose measurement takes 120 seconds to complete and the app displays the vital signs. The VitalDetect optimizes the signal quality by performing a signal quality check on the finger under measurement. If there is insufficient signal, the user must move the device to a finger with sufficient signal-to-noise ratio (SNR). Studies are currently been carried out at Pfützner Science and Health Institute (Mainz, Germany) in order to quantify the accuracy of the non-invasive glucose measurements in comparison to the YSI Stat 2300 glucose reference method.
Result:
In a first pilot experiment with 8 patients (N=5 type 2 and N=3 type 1) and employing a meal experiment, the VitalDetect UV signal showed a 65% change in the R = (AC/DC)UV/(AC/DC)IR signal with respect to the infrared signal, which went in parallel to the YSI reference values. The baseline noise in the R value was 12% implying that the observation might already be statistically significant. The studies shall conclude in September 2019 with 30 patients being tested after improvements to the algorithm. In the second experiment, four weeks later, with the same 8 patients, the baseline measurements (x2) were compared to the measurements for 3 patients after a meal (x2) and a R2 = 0.88 was found after using a two-point calibration with the strip method.
Conclusion:
Further work will now be performed to understand the UV response versus the IR signal between patients over the length of the meal experiment. It appears clear from the data that the effect of vascular perfusion, which is measured by the UV response, is also an indicator of general health and might be used to define a Vascular Health Index (VHI), which could be an indicator for the probability to develop type 2 diabetes.
Adsorption of Insulin to Infusion Lines is a Function of Flow Rate and is Clinically Significant
University of Canterbury Christchurch, Canterbury, New Zealand
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
Division of Endocrinology, Diabetes, Metabolism and Nutrition, Department of Internal Medicine, Mayo
Clinic Rochester, MN USA
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
Profil Neuss, Germany
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
Grand Canyon University, College of Science, Engineering and Technology
Phoenix, AZ, USA
Objective:
Detection of insulin could allow for tighter glycemic control by allowing patients to accurately dose their exogenous insulin. By developing a point of care (POC) sensor similar to current glucose test strips, patient compliance will not be hindered while their glycemic control is ameliorated. To properly develop a POC insulin sensor for daily use, different molecular recognition elements (MRE) must be evaluated. Prior work has validated the use of the insulin antibody for POC application whereas current work has shifted to designing the sensor for manufacturing as well as using an aptamer which shows promise for continuous sensing.
Method:
The antibody was immobilized on the surface of a gold screen-printed electrode using an activated self-assembled monolayer (16-Mercaptohexadecanoic acid). Additionally, to reduce the cost, carbon screen-printed electrodes are being explored using glutaraldehyde to crosslink the antibodies. However, as the aptamer is thiol modified, it can bind directly to the gold surface. After the MRE was immobilized, an insulin sample was applied to the sensor and the impedance was measured using electrochemical impedance spectroscopy.
Result:
The antibody-based gold screen-printed sensor’s calibration curve has an R-squared of 0.96 and a slope of 15.90 mOhm/pM. The aptamer’s calibration curve has an R-squared of 0.95 and slope of -0.93 mOhm/pM. Preliminary continuous testing showed increased insulin binding through increasing impedance response.
Conclusion:
The obtained data have validated the antibody and aptamer for POC use. However, the use of a carbon-based sensor does not have reproducible data yet. It also appears feasible to use the aptamer for continuous testing. Future work includes performing interference testing and continuous testing of the aptamer in both purified solution and using an animal model.
Fiasp® (Fast-Acting Insulin Aspart) Use with a MedtronicTM 670G System
Stanford University Stanford, CA, USA
Objective:
To assess the efficacy and safety of using Fiasp® in the hybrid closed-loop Medtronic™ 670G System.
Method:
This was a randomized, blinded, cross-over study of established users of the Medtronic™ 670G System. The first 2 weeks were an optimization phase with participants using their home insulin with weekly data review for adjustments. Following optimization, subjects were randomized to receive either insulin Novolog® or Fiasp® for 2 weeks, followed by the other insulin for the next 2 weeks. During the time of blinded wear, no additional optimizations were done. Allowing one week for 670G adaptation, data from the second week of blinded use of each insulin was used to compare the two insulins.
Result:
A total of 19 adult subjects were recruited, age 40±18 years, (10 male and 9 female) with a diabetes duration of 27±12 years. Results are given as mean ± SD or median (IQR). The baseline insulin use was 0.96±0.84 units/kg/day and A1c was 7.0±0.7%. Auto Mode use for Novolog® was 85.4±9.7% and 84.2±9.7% for Fiasp®. For Novolog® and Fiasp® respectively the percentage time in range (70-180 mg/dL) was 75.6 ±9.3% and 77.6 ±10.2%; % time <70mg/dL was 3.0 (1, 5)% and 2.0 (1, 3)%; mean glucose was 144.8 ±12.0 mg/dL and 147.2 ±13.5 mg/dL; coefficient of variation was 28.9 ±4.0 and 26.9 ±4.4; total daily dose was 40.8 (28.3, 64.3) units and 40.5 (32.5, 75.4) units; total daily basal was 17.6 (15.6, 31.1) units and 19.1 (15.9, 34.4) units. All differences were non-significant.
Conclusion:
Fiasp® is non-inferior to insulin Novolog® when used with an unmodified Medtronic™ 670G System. Further optimization of the insulin models used in this system might allow Fiasp® to achieve improved glycemic outcomes.
Biosensing of Diabetes Markers using Electroactive Aptamers based on Square Wave Voltammetry Principle
Joint Department of Biomedical Engineering, University of North Carolina at Chapel Hill and North
Carolina State University
Chapel Hill, NC, USA
Objective:
To realize the handheld POCT devices for various biomarkers for diabetes, such as vascular endothelial growth factor (VEGF) as the early diagnosis of diabetic kidney disease (DKD), vaspin for type 2 diabetes, and several glycated proteins, the development of novel molecular recognition elements which are suitable for electrochemical sensing is the key issue. Aptamers are DNA or RNA ligands that recognize a specific target molecule with comparable affinity and specificity to antibody. Electrochemical aptamer sensor has advantages of thermal stability, ease of chemical modification, and miniaturization. In this study, new redox probe modified aptamers and their electrochemical properties were investigated based on square wave voltammetry (SWV) for potential applications to develop aptamer-based electrochemical biosensing system for diabetes biomarkers.
Method:
The new redox probe, phenazine ethosulfate; 1-[3-(Succinimidyloxycarbonyl)propoxy]-5-ethylphenazinium triflate (amine reactive PES: arPES), modified VEGF aptamer was immobilized on the surface of a gold electrode. After PES aptamers were immobilized, VEGF samples were applied to the sensor and the signals based on SWV were monitored using different frequencies to optimize the detection.
Result:
The fabricated sensor showed electrochemical response upon incubation of VEGF, while there was no change with negative control, human serum albumin. This response was caused by a structural change to the aptamer enabling us to detect a target without bound/free separation. Not only VEGF, but biosensing of vaspin and glycated proteins will be also reported.
Conclusion:
The PES modified aptamer showed specific signals for diabetes biomarkers, in a washing-free manner, providing a new platform for the detection of diabetes-related biomarkers. Utilizing different aptamers, this principle is potentially universal for varieties of other biomarkers and therapeutic molecules for diabetes.
Reducing Size and Power Requirements for Long-Term, Needle-Implantable CGMs
Biorasis, Inc. Mansfield, CT, USA
Objective:
The objective of this work is to develop and perform feasibility studies on a highly miniaturized, needle-implantable biosensing platform that is capable of continuous glucose monitoring (CGM) using compact optical powering and communication methodology. The CGM system comprises glucose sensing elements outfitted with a custom application specific integrated circuit (ASIC) chip, fabricated using 65 nm CMOS technology geared to reduce both size of the implant and its power requirements.
Method:
The electrochemical glucose biosensors under development at Biorasis/UConn were integrated with 65 nm CMOS signal processing chip. This chip interacts with the glucose sensing electrodes and transmits digital pulses optically through the skin to the proximity communicator. The sensing elements were coated with dexamethasone-eluting biocompatible coatings that prevent foreign body response. The sensors were tested both in vitro (in PBS) and in vivo (in rat, rabbit and mini-pig animal models). Blood glucose concentrations were used as a reference. Biorasis’ custom printed circuit board (PCB) housed inside the proximity communicator was used to collect, process, and report/store glucose data. Prototypes were tested in vitro for response to glucose. The platform has the flexibility to perform a number of complex functions including sensing, alerts and internal calibration routines.
Result:
The CGM system showed a linear response to glucose in vitro throughout the physiologic glucose range (2.5 – 25 mM). The custom PCB for the proximity communicator was optimized to be able to alter its sampling frequency, and perform first-stage calibration routines.
Conclusion:
A needle-implantable CGM system outfitted with a custom ASIC 65nm CMOS chip was developed and successfully tested to in vitro to measure glucose level. The data collected so far prove feasibility of this approach. This long-term, miniaturized, needle implantable CGM system is currently being adapted for clinical studies under the support of Helmsley Charitable Trust.
Software Tool for Glucose Monitoring Data Processing in Diabetes Studies
Universidad Antonio Nariño Bogotá, Colombia
Objective:
To develop a software tool for glucose monitoring data processing that supports analysis metrics for diabetes research studies.
Method:
Implement a set of typical diabetes metrics of analysis through Matlab code to simplify the pre-processing and processing of glucose monitoring data. Differences in data files from Medtronic's systems of continuous glucose monitoring and insulin pumps were considered for importing processes and data debugging.
Result:
A software tool in development at Matlab has been used in several diabetes studies facilitating the data processing from continuous glucose monitoring systems and insulin pumps. So far, most studies where it was used have been focused on glucose variability analysis, glucose control performance of sensor augmented pump therapy, metabolic control, and risk of hypoglycemia for several type 1 and type 2 diabetes research studies. Metrics implemented include: MAGE, MAG, Mean, Std, CV, CONGA, LBGI, HBGI, ADDR, IQ range, MODD, AUC, hypoglycemic ranges, hyperglycemic ranges, time percent in hypo- and hyperglycemic ranges, M value, J index, hypo- and hyperglycemic events for different thresholds, suspension duration, glucose in suspensions, etc. All of them obtained also by day, day-time, or night-time.
Conclusion:
The software tool was used to support analysis of diabetes research studies in an easy and adaptive way according to particular needs of the clinical team for each study. Several clinical studies were boosted from this software tool development allowing to support additional diabetes studies achieving a good cohesion between the clinical and engineering scope in a reputed Hospital of Bogotá, Colombia.
Use of a Closed Loop Automated Insulin Delivery System in Veterans Over 65 Years
VA Northeast Ohio Healthcare System Department of Pharmacy Cleveland, OH, USA
Objective:
Insulin pump therapy and continuous glucose monitoring present many challenges. The Medtronic MiniMed 670G system, a hybrid closed-loop (HCL) system, adjusts basal insulin in real time to keep glucose levels at 120mg/dl. Use of this system increases time in range, while reducing hypoglycemia and A1c. This device has been studied in diverse groups of patients but potential benefit for patients over 65 years has not been confirmed. The endocrine team at the VA Northeast Ohio Healthcare System (VANEOHS) has transitioned 56 patients to this system since it became available in 2017. The purpose of this study is to evaluate the performance and acceptability of this technology for veterans in this age group.
Method:
Retrospective chart review of VANEOHS patients over 65 years of age utilizing the HCL insulin delivery system.
Result:
A total of 20 patients over 65 years old were identified. Eighteen of the 20 patients (90%) utilized ‘automode’ for the majority of the time. An average decrease in A1c of 0.51% was observed after an average of 10 months on device. For patients in ‘automode’, time in range was 74% which exceeded the national average of 71.6% reported by the device manufacturer. Hypoglycemia in our population (defined by time with glucose <55mg/dl) was 0.2%, which compared favorably with the national average of 0.5%. Overnight time in range and rates of nocturnal hypoglycemia also improved in this cohort.
Conclusion:
A closed loop automated insulin delivery system demonstrated similar benefit to veterans over 65 years as in a more diverse patient population. Veterans at the VANEOHS exceeded expectations for time in range and rates of hypoglycemia. HCL insulin delivery systems are a viable option for seniors with diabetes.
Deep Transcranial Magnetic Stimulation for the treatment of Type 2 Diabetes
Endocrinology and Metabolism Division, IRCCS Policlinico San Donato San Donato Milanese, Milan, Italy
Objective:
We recently demonstrated that Deep Transcranial Magnetic Stimulation (dTMS) is capable of inducing satiety and body weight loss in obesity, through modulation of dopaminergic system. The main objective of this study is to evaluate the efficacy of dTMS in the control of food craving and in the reduction of body weight in diabetic and obese subjects. The effect of dTMS on metabolic parameters related to glucose metabolism (glycated hemoglobin and fructosamine) is also evaluated.
Method:
A total of 36 obese patients (10 males, 26 females, age 48.4±9.9 yrs, BMI 36.4±4.4 Kg/m2) were randomized and completed the study: N=7 obese diabetic (DO) and N=15 non-diabetic obese (NDO) subjects underwent a 5-week treatment with high-frequency (HF, 18 Hz) dTMS and N=4 DO and N=10 NDO subjects were Sham-treated. Food craving and metabolic parameters were evaluated at baseline, after the 5-week treatment, and at follow-up visits (1 month (FU1), 6 months (FU2), 1 year (FU3), after the end of the treatment).
Result:
The weight reduction was significantly higher in the DO18Hz group than in DOsham (respectively -3.9±1.2% and - 1.4±1.3%; p=0.022). A significant reduction in glycated hemoglobin was observed in DO18Hz group: after 5-weeks treatment (-8.4±3.0% vs baseline, p=0.015) and at FU1 (-9.0±2.8% vs baseline, p=0.010). A significant reduction in fructosamine was observed in DO18Hz group at FU1 (-7.5±4.3% vs baseline, p=0.009).
Conclusion:
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
Objective:
A new prandial insulin strategy was investigated on the hybrid closed-loop (HCL) systems to observe the potential of freeing patients from carbohydrate counting while maintaining adequate glycemic control. The sensitivity tests on carbohydrate size approximations were examined as well.
Method:
We integrated the UVA HCL system with a new meal control module, combining temporarily increased basal rate and a priming bolus (both computed from total daily basal insulin, and triggered by meal announcement without carbohydrate counting). This new method was compared to standard HCL (using carbohydrate counting and carbohydrate:insulin ratios) using the UVA/Padova T1D simulator. Single meals of varying size were simulated in 100 virtual adults, from a variety of fasting glycemic states. Glycemic control was assessed by computing time spent between 70mg/dL and 180mg/dL and time spent below 70mg/dL over the prandial excursion. The sensitivity tests were performed on a one-day scenario, with or without carb counting errors (50%-150%). There were three meals (breakfast, lunch, and dinner) and an afternoon snack, given the pre-assumed probabilities of carbohydrate size misrecognition.
Result:
The new carbohydrate independent strategy achieved similar performance (time-in-range), and better protection against hypoglycemia when compared with HCL across the different meal sizes. In sensitivity tests, the differences of time-in-range and time-in-hypo did not exceed 3 percent of the overall time, respectively, between the groups of correct and misrecognized carbohydrate sizes.
Conclusion:
A new bolusing strategy for HCL, without carbohydrates counting, was shown in-silico to have similar performances and an improved safety profile when compared with standard HCL. Sensitivity analysis showed stable performances across a broad range of meal sizes and compared favorably to standard bolusing in the presence of counting errors.
The Effect of Glucose Scanning Frequency on Glycemic Control in Individuals with Type 1 Diabetes using Flash Glucose Monitoring
Medical University of Graz Graz, Styria, Austria
Objective:
Monitoring glucose using continuous glucose monitoring (CGM) has shown to improve A1c in individuals with type 1 diabetes on either multiple daily injections (MDI) or continuous subcutaneous insulin infusion (CSII) therapy across all age groups. During the last few years, flash glucose monitoring (FGM) has become widely available in many countries. This study aimed to examine the effect of glucose scanning frequency on A1c in individuals with type 1 diabetes using FGM.
Method:
We evaluated the effect of scanning frequency using FGM (≤7 vs. >7 scans per day) in 95 individuals on MDI or CGM therapy (47 females and 48 males, age 42.2 ±14.1 years, BMI 25.2 ± 4.0 kg/m2, diabetes duration 20.8 ± 13.1 years, A1c 56.8 ± 9.2 mmol/mol) over a three month period. Data from a tertiary healthcare center registry were used for this analysis.
Result:
A1c significantly improved with increased scanning frequency (≤7 scans: 61.5 [54.5-65.6] vs. >7: 54 [49.0-61.0] mmol/mol; p=0.02). A similar effect was seen for the percentage of time in target range (70-180 mg/dl) (≤7: 45.0 [38.0-51.0] vs. >7: 57.0 [48.0-65.0] %; p=0.009). No effect on the percentage of time spent below range was observed with increased scanning frequency (≤7: 6.0 [4.0-8.0] vs. >7: 5.0 [3.0-8.0] %; p=0.258].
Conclusion:
Scanning glucose values with an FGM system more than 7 times per day improves A1c and increases time in target range in individuals with type 1 diabetes on MDI or CSII therapy. No effect was seen in time spent in hypoglycemia but this might be attributed to the low rate of hypoglycemia in the investigated population.
Retrospective Study of Inpatient Diabetes Management Service, Length of Stay and 30-Day Readmission Rate of Patients with Diabetes at a Community Hospital
New York University New York, NY, USA
Background:
Hospitalized patients with diabetes are at risk of complications and longer length of stay (LOS). Inpatient Diabetes Management Services (IDMS) are known to be beneficial; however, their impact on patient care measures in community, non-teaching hospitals, is unknown. The objective of this study is to evaluate whether co-managing patients with diabetes by the IDMS team reduces LOS and 30-day readmission rate (30DR).
Method:
This retrospective quality improvement cohort study analyzed LOS and 30DR among patients with diabetes admitted to a community hospital. The IDMS medical team consisted of an endocrinologist, nurse practitioner, and diabetes educator. The comparison group consisted of hospitalized patients with diabetes under standard care of attending physicians (mostly internal medicine-trained hospitalists). The relationship between study groups and outcome variables was assessed using Generalized Estimating Equation models.
Result:
A total of 4,654 patients with diabetes (70.8 ± 0.2 years old) were admitted between January 2016 and May 2017. The IDMS team co-managed 18.3% of patients, mostly with higher severity of illness scores (p < 0.0001). Mean LOS in patients co-managed by the IDMS team decreased by 27%. Median LOS decreased over time in the IDMS group (p = 0.046), while no significant decrease was seen in the comparison group. Mean 30DR in patients co-managed by the IDMS decreased by 10.71%. Median 30DR decreased among patients co-managed by the IDMS (p = 0.048).
Conclusion:
In a community hospital setting, LOS and 30DR significantly decreased in patients co-managed by a specialized diabetes team. These changes may be translated into considerable cost savings.
Real-World Patient Experience from 5,024 Patients Using the t:slim X2™ Insulin Pump with Basal-IQ® Technology
Tandem Diabetes Care San Diego, CA, USA
Objective:
The t:slim X2™ insulin pump with Basal-IQ® technology predicts glucose levels 30 minutes ahead and suspends insulin to help reduce the time and frequency of low-glucose events. This is the first large scale study to evaluate patient reported outcomes (PROs) from Basal-IQ technology users during real world use. Patient experience is an important indicator in the continued use of therapy and improved glycemic and psychosocial outcomes.
Method:
In May 2019, 5,024 users of the t:slim X2™ insulin pump with Basal-IQ® technology voluntarily responded to an online survey (Mage = 39; prior therapy = 60% pump, 40% MDI; 92% T1D; 53% female). Nine metrics measured satisfaction, trust, usability, and utility of the system on a 5-point Likert scale.
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
John A. Burns School of Medicine, University of Hawaii at Manoa
Kailua, HI, USA
Objective:
Continuous glucose monitoring (CGM) is increasingly utilized to improve and individualize diabetes patient insulin management using trend arrows, summary graphics, and statistics. Carbohydrate counting (CC) remains the standard of care in determining insulin dosing. However, protein and fat contribute to the height, shape, and duration of the CGM glucose concentration curves, as does timing and dosing of insulin administration. Here, calculated Glucose Clearance (GLCL) is used to assess the insulin effect on usual mixed-meal appearance/disposal to improve diabetes management with insulin in a type 2 patient.
Methods:
Calculating GLCL requires kilogram body weight and an estimate of Volume of Distribution (Vd): 0.18 * kg body weight*10; GLCL = Vd/t, (t = time of glucose appearance/return to baseline. GLCL is then related to the food components and size of the meal, along with the pre-meal timing and amount of insulin injected. Post-meal activity may also be noted.
Results:
Meals as examples: 1) Breakfast: Oatmeal (1 cup), strawberries (6), ½ cup blueberries and ½ cup cottage cheese; Weight = 77.9 kg, 13 units Humalog 20 minutes before the meal. Peak glucose 180 mg/dL. Vd = 140 dL; time = 210 minutes; GLCL = 66.7 mL/minute. Insulin impact; 66.7/13 = 5.13 mL/(min*unit). 2) Lunch: 1.5 cups pasta with grated cheese and zucchini, 18 units Humalog 35 minutes pre-meal. Time = 300 minutes. Peak glucose 185 mg/dL. GLCL =46.7 mL/min; Insulin impact: 46.7/18 = 5.5 mL/(min*unit).
Conclusion:
Insulin effect with usual meals can be quantitatively assessed without carbohydrate counting using CGM data. In the examples above, despite differences in meal composition and insulin amounts, GLCL/unit insulin values were comparable and would serve as minimal insulin administered for a repeat same meal.
Glycemic Control Assessment based on Consensus CGM Metrics using less than 14 Days
Novo Nordisk A/S Bagsværd, Copenhagen, Denmark
Objective:
Continuous glucose monitoring (CGM) technology can improve outcomes for diabetes patients. An expert panel has recently reached consensus on clinical targets for CGM data interpretation. A total of 14 days of CGM is recommended to assess glycemic control. However, in clinical practice less CGM data may be available. The aim of this study was to explore the ability of the recommended metrics to assess glycemic control using less than 14 days of CGM data.
Method:
CGM data was acquired from Cornerstone4Care powered by Glooko. Data from 222 DM patients were included in this study (N=1320 14-day profiles). All calculations were based on 14-day periods with >70% of possible CGM readings. The standardized CGM metrics were calculated for each patient. The percentage deviations of the metrics were found between each accumulated day (1 to 13) and the accumulated 14 days. The ability of the metrics to assess glycemic control using less than 14 days of CGM data was examined by comparing whether all metrics indicated acceptable glycemic control for each accumulated day (1 to 13) compared to the fully accumulated 14 days.
Result:
For all the recommended metrics, a decrease in percentage deviation was observed with increasing CGM days. Similarly, the ability to assess glycemic control improved in terms of accuracy by increasing CGM days considering the guidance on targets based on the provided consensus.
Conclusion:
The more CGM days included, the better is the resemblance to the metrics obtained from the 14 days. However, valuable insights about a patient’s glycemic control can be obtained, even in situations where less CGM data than recommended are available.
SARIMA Local Modelling for Online Glucose Prediction
Polytechnic University of Valencia València, Spain
Objective:
The use of WLM-SARIMA modeling framework (1.fuzzy clustering algorithm (FCM) used for finding similar glycemic responses, 2.stochastic seasonal SARIMA model obtained for each cluster, 3.glucose prediction computed by weighting SARIMA local models following memberships (M) cluster structure) has shown very good results for long prediction horizons (PH). However, weights online computation is an open issue.
Method:
Long-term simulated data (6 months, 80% training/20% validation) were generated, and a WLM-SARIMA model identified from “event-to-event” time subseries (enforcing seasonality) by combining FCM with a SARIMA local model for each cluster. Online M estimations are calculated by computing cluster prototypes to last four (15min.) continuous glucose monitor (CGM) data distance. Glucose predictions are then calculated by combining SARIMA local estimations with M. Additionally, a “reliability” prediction measure (R) is calculated based on the M second derivatives (convexity).
Result:
The WLM-SARIMA approach achieved the following mean absolute percentage errors (%): 5.95, 8.43, 11.32, 13.65, and 13.97 (for PH of 30, 60, 120, 180, and 240 minutes, respectively), similar to the off-line results found in previous work (3.14, 6.37, 11.93, 14.10 and 15.45). R gives values near to one when a model similar to recent (15min.) CGM data are available and very low when many models must be combined to give good estimations, alerting the user of the lack of adequate models for this particular behavior.
Conclusion:
Online use of WLM-SARIMA is possible with the proposed M estimation procedure and the new R gives additional information about the certainty of forecasted values. Future work in this project funded by MINECO DPI2016-78831-C2-1-R includes the online creation of new clusters and local models when new behaviors are detected by R low values.
Removal of Phenols from Insulin Suppresses Inflammation & Promotes Blood Glucose Regulation In Vivo
Wayne State University Detroit, MI, USA
Objective:
Blood glucose regulation (BGR) via subcutaneous insulin infusion frequently fails within the first 2-3 days of infusion. Commercial insulin formulations contain phenol and/or m-cresol (P/C), which serve to stabilize insulin in vitro, but are known to be cell and tissue toxic, leading to inflammation and tissue destruction at the infusion sites. We hypothesize that in-line removal resins (exchange/filtration) can remove P/C without loss of functional insulin and BGR.
Method:
P/C is removed from commercial insulin formulations (Humalog) and diluents using Zeolite-Y (Z-Y), a size exclusion/ion exchange based resin. P/C levels are determined post Z-Y filtration using HPLC/PDA. Cytotoxicity is assessed pre- and post-filtration of Humalog and diluent in vitro whereas a modified murine “air pouch” model was used to assess toxicity in vivo.
Result:
Our studies demonstrated that Z-Y effectively removed P/C from Humalog without removing insulin. In vitro studies demonstrated that removing P/C significantly decreased cell toxicity of Humalog. In vivo studies demonstrated that insulin infused through an in-line Z-Y filter for CSII has an equivalent capacity for BGR in diabetic mice. Histological analysis of infusion sites demonstrate leukocyte accumulation in the pre-filtered insulin infused tissue site, but not in saline or zeolite filtered insulin tissue sites.
Conclusion:
These studies provide evidence that the P/C present in commercial insulin formulations induce inflammatory reactions at insulin infusion sites. These studies support the effectiveness of the removal of P/C prior to insulin infusion in minimizing the loss of insulin function, as well as insulin destructive inflammation at CSII infusion sites.
Effect of Using Multiple Sensors on Activity Classification Performance and Smartwatch Battery Life
Rensselaer Polytechnic Institute Troy, NY, USA
Objective:
Sixty-seven percent of adolescent patients with type 1 diabetes forget to bolus for ≥1 meals/week, causing suboptimal glucose control. Sixteen percent of patients encounter hypoglycemic events even with perfect exercise detection and stopping insulin. Automatic activity detection using smart devices can assist such people by prompting meal announcements and anticipating exercise, however, such a system must be practical to use. This work evaluates the tradeoff between activity classification performance and smartwatch battery life across sensors and Android smartwatches.
Method:
Smartwatch and smartphone data are used for minute-based classification of eating, exercise, sleep, other, and null (inactive device) activities. 2-D convolutional neural networks (CNN), Long Short-Term Memory (LSTM), and probabilistic graphical models (PGM) were used to classify watch accelerometer and gyroscope data. An LSTM was used to classify phone accelerometer data. A lookup table of previous locations was used to classify a given location. A single layer neural network was used for sensor fusion. Minute based outputs were filtered using a PGM to detect activity events. Smartwatch battery life was optimized by using sensor buffers and periodically saving data.
Result:
The CNN achieves the highest accuracy of 85% given watch accelerometer and gyroscope readings. However, the CNN is 84% accurate when given only the accelerometer. Sampling just the accelerometer on a Motorola 360 smartwatch extends battery life by ~12 hours. The accuracies of the phone accelerometer LSTM and location lookup are 48% and 61% respectively. Combining location and phone accelerometer outputs to watch sensor outputs only slightly improves classification accuracy. A >95% event detection accuracy is achieved for sleep, eating, and exercise.
Conclusion:
Based on classification accuracy, the smartwatch is the most useful device for activity detection. Phone accelerometer and location should be used only when the watch is unavailable. Sampling only the watch accelerometer sacrifices little accuracy and significantly extends battery life.
Pilot Study for Non-Invasive Glucose Monitoring by means of Photoplethysmography
Sanmina Huntsville, Alabama
Background:
For non-invasive assessment of glucose, endothelial vascular function, and other body parameters, a new device was developed, which includes an optical circuit configured to detect several photoplethysmography (PPG) signals. The first PPG signal includes a first spectral response obtained from light reflected around a first wavelength, and a second PPG signal includes a second spectral response obtained from light reflected around a second wavelength from the tissue of the user.
Method:
This technology was investigated in an IRB approved pilot study, which served the primary purpose to improve the underlying algorithm used to extract the glucose concentration from the obtained readings.
Result:
A total of 12 patients participated in the trial (N=5 female, N=7 male, N=5 type 1, 7 type 2 diabetes, age: 57±18 yrs). They received a standardized meal and glucose was measured at 11 time-points over a period of 3 hrs. YSI 2300Statplus served as reference method for the non-invasive readings. The observed measurement range was 59 mg/dL to 371 mg/dL. The device worked well in all patients with one exception (a male patient with very thick skin). All results were included into the analysis. Mean absolute relative difference over the entire data set (n = 104) was 8.0 %. In the consensus error grid, 98 % of the results were in zone A and 1.9 % were found in zone B.
Conclusion:
In conclusion, the prototypes tested were employing a new and promising non-invasive glucose assessment technology, which was shown to reliably and accurately measure glucose levels in this first pilot study.
Machine Learning Approaches to Enhance Insulin Bolus Dosing in Type 1 Diabetes Therapy
Department of Information Engineering, University of Padova
Padova, Italy
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)
Division of Endocrinology, Icahn School of Medicine at Mount Sinai New York City, NY, USA
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
diafyt MedTech, a pg40 Consulting Group company Leipzig, Saxonia, Germany
Objective:
Determine the effectiveness of insulin dosage calculation by self-learning systems to optimize glucose levels in type 1 diabetics.
Method:
Insulin dosages from preclinical data were calculated by experts and the device algorithm, compared, and evaluated. The "knowing observer", the expert can see the future outcome (future data), with the assumption that with this knowledge, the expert can determine the right amount of insulin. The algorithm in contrast will calculate the insulin dosage only with information at this point in time and before as same as it would be in reality.
n=1000
first expert: k1=n, Practitioner
second and 3rd expert: Random sample of 30 control values will satisfy a significance level of 5%, k2,3=30, Medical Doctor
Result:
Data show a strong correlation between machine-calculated insulin demand and the expert insulin demand determination.
Conclusion:
Preclinical tests show promising results for effective insulin dosage calculation by self-learning systems
EGP Variation in First 24 Hours of a Critically Ill New Zealand SPRINT Cohort
University of Canterbury, Dept of Mechanical Engineering, Centre for Bioengineering Christchurch, Canterbury, NZ
Objective:
Control of hyperglycemia in critically ill patients using glycemic control (GC) can improve patient outcomes. Endogenous glucose production (EGP) can be a significant contributor to the overall modeled glucose flux and model-based insulin sensitivity (SI) used to guide GC, and is highly variable in the first 24 hours of stress response. Using a constant EGP value provides good model fit for most patients, but some patients fall outside these bounds due to extreme stress response. This study assesses occurrence of extreme EGP and for what percentage of hours.
Method:
Data from a New Zealand cohort of critically ill patients (N=202) that used the Specialized Relative Insulin Nutrition Tables (SPRINT) GC protocol were analyzed. Patients were on SPRINT for at least 24 hours and started the protocol within 12 hours of ICU admission. The ICING model was fit to patient data for EGP from 1.16-2.5mmol/min. Low EGP was defined as an SI constrained to its minimum and blood glucose error was over 0.1mmol/L. Increasing EGP to obtain this minimum fit defines a minimum value of elevated EGP, and the number of patients and hours where such elevated stress response occurs is quantified.
Result:
When EGP was 1.16mmol/min, 42.6% of patients (N=86) had at least one hour of low EGP. As EGP increased, the number of patients with a constrained value decreased, and model-fit error decreased by 16.3% and 8.9% respectively when EGP was 1.25mmol/min and 1.5mmol/min. An EGP value of 2.50mmol/min corresponds to 0.5% of patients (N=1 patient and 1 hour) with a constrained value, or 0.023% of patient hours. The constrained values are concentrated in the first 0-12 hours on the SPRINT protocol.
Conclusion:
This study quantifies the numbers of patients and hours in which significantly elevated stress response can occur in the first 24 hours of ICU stay for those patients on GC. These hours also quantify the likelihood or risk in providing GC for those patients.
Representation of Type 1 Diabetes (T1D) Device Data using CDISC Clinical Data Standards
CDISC Austin, Texas, USA
Objective:
Revolutionary developments in technology are enabling more effective management of diabetes. Creation and adoption of clinical data standards supporting diabetes technology data will transform incompatible and disparate data into universal and illuminating information, facilitating discoveries that could have invaluable impact on type 1 diabetes (T1D) clinical research. Implementation of Clinical Data Interchange Standards Consortium (CDISC) standards allow collection, organization and analysis of data in a clear and consistent manner enabling all researchers to leverage information from studies globally. CDISC standards enable the accessibility, interoperability, and reusability of data, driving operational efficiencies, expediting regulatory review, and reducing time to market.
Method:
With support from The Leona M. and Harry B. Helmsley Charitable Trust, CDISC is leading a unique, consensus driven effort, bringing together T1D experts from academia and industry to create T1D clinical data standards in
Result:
Project scoping identified the following key areas for development in diabetes technology:
- 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
- Exercise Activity types and Context
- Fitness and Strength Status
- Meal Descriptions and Nutrient Content
Conclusion:
Key diabetes technology concepts were identified for creation of CDISC standards. After the development process, the T1D community is encouraged to participate in the informational webinars and public review stages, which will enable all-inclusive development, approval, and adoption of the data standards.
DietHabit: Understanding Dietary Habits in Type 2 Diabetes from Electronic Food Diaries
Rice University Houston, TX, USA
Objective:
Diet plays a significant role in self-management of diabetes and dietary habits are often challenging to change. While there is a basic belief that dietary choices are repetitive due to acquired food and taste preferences, there exists no quantified understanding of “dietary habits.” We aim to find mathematical representations of diet that can enable the development of generalizable diet habit models to establish individual food habits and create intervention-specific personalized food dictionaries for better diet management.
Method:
We analyzed a diabetes dataset that comprised of dietary logs over 6 months from N=10 type 2 diabetes participants in a lifestyle intervention study. Each subject described similar dishes very differently according to their choice of labels. So, it was challenging to classify different records as similar or dissimilar across participants. Thus, we proposed to extract fixed-length mathematical representations of each logged food item using word2vec representations and a Long short-term memory (LSTM) auto-encoder, with a goal to establish a closeness/distance measure between entries across individuals to develop a model that represents their dietary habit.
Result:
To have a generalizable representation, we compared similar dishes described differently by study participants and established closeness measures for each dish. The fixed length mathematical representations provided an excellent framework to compare textually described food dishes quantitatively. Analysis with a cosine similarity metric led to a clustering of dishes that were described differently but were very similar.
Conclusion:
We presented diet habit analysis on a diabetes diet log dataset to understand dietary choices quantitatively and to establish intervention specific personalized food dictionaries that can improve participant compliance in diet interventions which are crucial for diabetes management.
Behavior Problems Predict Self-Monitoring Blood Glucose Frequency in Children with Recent-Onset Type 1 Diabetes
University of Kansas Medical Center Kansas City, KS, USA
Objective:
Research suggests that child problem behaviors negatively impact frequency of self-monitoring of blood glucose (SMBG) in children with established type 1 diabetes (T1D; >12months); however, no studies have examined this relationship during the T1D recent-onset period (≤12months). We examined the impact of child problem behaviors on SMBG frequency trajectories in a cohort of families of children with recent-onset T1D.
Method:
Parents completed the Eyberg Child Behavior Inventory (ECBI) and a demographic survey at baseline. We downloaded child glucometers at baseline, 3-, 6-, 9-, and 12-month assessments and used these data to calculate mean daily SMBG for each timepoint. We conducted group-based trajectory modeling (GBTM) to identify and characterize 12-month SMBG trajectories and used multinomial logistic regression to examine baseline child problem behavior as a predictor of trajectory membership.
Result:
Seventy-eight families completed baseline and follow-up assessments (Mage=7.44yrs, MDXduration=4.36mos, 47.4% male, 88.2% non-Hispanic, White). Mean SMBG ranged from 5.98-6.40 checks per day across assessments and 19.2% of parents reported clinically elevated ECBI scores at baseline. GBTM showed a 3-group trajectory solution: group1 showed decreasing daily SMBG (baseline SMBG=4.8, 32.7% of sample), group2 showed stable SMBG (baseline SMBG=6.5, 48.2% of sample), and group3 showed increasing SMBG (baseline SMBG=8.3, 19.1% of sample). Controlling for family income, ECBI scores predicted trajectory group membership, with higher ECBI scores associated with membership in the group with decreasing daily SMBG (b=-0.17, p=0.02).
Conclusion:
Children with elevated behavior problems soon after T1D diagnosis may be at increased risk for poor and declining SMBG frequency. Screening for child problem behaviors after diagnosis and preventative interventions (e.g., behavior management) may help to maintain higher SMBG frequency during the recent-onset period.
The Mealtime Insulin BOLUS Score Increases Prior to Clinic Visits in Youth with Type 1 Diabetes
University of Kansas Medical Center Kansas City, KS, USA
Objective:
Studies confirm the occurrence of “White Coat Adherence,” a term describing an increase in self-care engagement just prior to a clinic appointment, in multiple illness populations. In youth with type 1 diabetes (T1D), research also shows an increase in self-monitoring blood glucose frequency (SMBG) ahead of the child’s clinic visit. Here, we extend the literature to determine if white coat adherence also occurs with mealtime insulin bolusing (BOLUS) in youth with T1D.
Method:
We extracted insulin pump records and A1c levels from a clinical database of youth with T1D. We calculated a mean BOLUS (maximum of 1 point/meal or 3 points/day) for 6-5 weeks, 4-3 weeks, and 2-0 weeks prior to youths’ routine clinic visits. We used multilevel modeling to examine patterns of BOLUS scores prior to clinic visits and examined for age differences.
Result:
We extracted data for 459 youth (Mage=12.5±2.9yrs, MDXduration=4.6±3.5yrs, MA1c=8.4±1.6%). Mean BOLUS scores ranged from 2.2-2.4 across assessments. Multilevel modeling showed a significant increase in BOLUS scores in the weeks leading up to youths’ clinic appointments (b=0.07, p≤0.001). On average, adolescents had lower BOLUS scores than school-age children (b=-0.35, p≤0.001). Post-hoc analyses showed that adolescents consistently had lower BOLUS scores than children across assessments (p’s≤0.001).
Conclusion:
Families who increase their mealtime insulin use just prior to clinic appointments may help their youth achieve tighter blood glucose levels and reduce their youth’s risk of complications even if for a few weeks. However, T1D care teams and researchers may need to download insulin pump data for up to 6 weeks prior to a youth’s clinic visit to observe a youth’s typical level of mealtime insulin use.
A Needle Array and its Application in CGM Sensors
Zimmer and Peacock Royston, Hertfordshire, UK
Objective:
The objective was to develop a needle array capable of piercing the dermis. The needles were functionalized to detect glucose within the interstitial fluid. The intent of the research was to make available a design of needle array that were functionalized with glucose oxidase, and where the detection methodology was electrochemical amperometry.
Method:
The needle arrays were fabricated from a low-cost polymer. The needles were appropriately metalized and subsequently, a barrier layer, an enzyme layer, and a top coating were adhered onto of the electrodes.
Result:
The needle array was functionally tested and showed a good response to glucose challenges and were validated for exploitation within the continuous glucose monitoring (CGM) application.
Conclusion:
The conclusion is a program-ready needle array platform for application in CGM.
Clinical Evaluation of Acute Interference for a Combined Invasive and Non-Invasive Glucose Meter
Pfützner Science & Health Institute Mainz, Germany
Background:
This clinical study was conducted to investigate the potential acute impact of common nutritional substances, frequently used over-the-counter drugs, and popular nutritional supplements on the accuracy of the invasive and the non-invasive modules of the TensorTip Combo glucose meter (CoG, CNOGA Medical, Cesarea, Israel).
Method:
The ten healthy subjects included in this trial (N=6 male, N=4 female, age: 41±14 yrs., BMI: 26.5±5.4 kg/m²), received a personal CoG device and performed a three-day calibration procedure. During the following 10 visits, they arrived in the morning after an overnight fast and ingested a high dose of the respective test substance. Six blood samples for glucose assessment were drawn at time-points related to the anticipated pharmacokinetic profiles. YSI Stat2300 plus and COBAS served as reference methods. Mean absolute relative bias for each concentration was calculated and plotted against the plasma substance concentration. Interference was assumed when the slope of the regression line was >10% or <-10%.
Result:
For the non-invasive module, no interference was seen with any of the tested substances (i.e., acetaminophen, acetyl salicylic acid, ascorbic acid, caffeine, diclofenac, ethyl alcohol, ibuprofen, mannose, xylose, 3Ω-fatty acids). Uptake of ethyl alcohol (0.2 g/kg) caused interference with the YSI results (-11 % vs. COBAS) and the invasive device module results (-11 % vs COBAS). Overall mean absolute relative difference (MARD) in the tested glucose range (71 to 158 mg/dL) was 8.8% and 7.4% for the non-invasive and the invasive module, respectively.
Conclusion:
In conclusion, the non-invasive CoG module was not influenced by any of the tested substances. Ethyl alcohol sample interference for the CoG module or the YSI device will result in low-biased results, but the decrease is not clinically relevant unless the user is completely drunk (e.g. plasma alcohol levels of 2.0 o/oo result in a 20 % underestimation by YSI or by the invasive CoG module.
De-Escalation Therapy (DET) – An Intensive Short-term Temporary Pharmacological Intervention to Stop Disease Progression in Patients with Type 2 Diabetes
Pfützner Science & Health Institute Mainz, Germany
Background:
Type 2 diabetes is a chronic progressive disease, naturally requiring continuous treatment intensification, and resulting in secondary microvascular and macrovascular complications. The underlying pathophysiological disorders (ß-cell dysfunction (ßCD), insulin resistance (IR), and chronic systemic inflammation (CSI)) are present in the individual patient with different degrees of severity. We have developed an alternative approach to the common drug escalation programs: intensive short-term multi-pharmacotherapy interventions (DET, de-escalation treatment) to achieve sustained improvement in ß-Cell function and to reverse disease progression.
Method:
The personalized treatment selection is based on the results of a biomarker panel consisting of classic biomarkers (glucose, lipids, A1c) and additional pathophysiology-oriented biomarkers (e.g. hsCRP, intact proinsulin, adiponectin). Here we report on the observational results (with up to 12 years of follow-up) of 22 patients (N=8 women, N=14 men, age: 62±8 yrs., disease duration: 12±7 yrs., A1c: 7.8 %, BMI: 33.2±2.4 kg/m²). The temporary intensive treatment approach lasted 3 months and consisted in general of low doses each of basal insulin to address ßCD, treatment of IR (exercise and/or pioglitazone), a component to reduce CSI (diet, GLP-1, or SGLT-II), and an additional hypoglycemic intervention (metformin, DPPIV), if applicable.
Result:
The DET approach was well tolerated (nausea: N=4 cases, edema: N=1 case). All patients experienced either a normalization or pronounced improvement of their diabetes control without any report of hypoglycemia. Mean A1c after 3 months improved to 5.9±0.4 %. Intact proinsulin decreased from 9.3±2.1 pmol/L to 2.3±0.6 pmol/L, adiponectin improved from 3.4±1.2 to 8.6±2.4 mg/dL, and mean body weight decreased by 2.4±1.1 kg (all: p<0.01). After the DET, the majority of the patients were continued with measures of either lifestyle only or a drug monotherapy targeting the major component of their individual diabetes phenotype. The DET effect lasted on average for about 2 years (range until next DET: 6 months to >11 years). Maintenance of the DET effect was associated with adherence to recommended lifestyle measures. Lack of compliance with the complex treatment requirements were the reasons for treatment failure in two patients resulting in the well-known chronic disease progression.
Conclusion:
Regeneration of ß-cell function by means of DET resulted in consecutive “type 2 honeymoon periods”, which helped to achieve a temporary stop of chronic disease progression. We are now conducting a formal prospective clinical study to investigate the practicability of this approach in routine clinical practice.
Miniaturization of an Osmotic Pressure-Based Glucose Sensor by Means of Nanotechnology Applications
Pfützner Science & Health Institute Mainz, Germany
Background:
The Sencell Glucose Sensor (Lifecare, Norway) uses osmotic pressure differences between a reagent chamber, containing an active fluid with a glucose binding molecule (GBM) and a glucose-like ligand, and the interstitial fluid, to determine interstitial glucose concentrations. The reaction is solely based on a reversible glucose binding to a specific receptor and no molecule is destroyed to generate the signal. This guarantees a long-term survival of the sensor in the body. Successful proof-of-concept studies have been performed in pigs with wired prototypes (2 x 1.5 x. 0.6 cm³).
Method:
To develop an injectable device, the core sensor technology needs to be significantly miniaturized without losing pressure sensing sensitivity. A 3D-printed nanosensor technology (cantiMED, Darmstadt, Germany) was employed to achieve this task. In a pilot approach, MEMS technology was used to build a sensor chamber with eight-times smaller dimensions than the previous version and the pressure membrane was equipped with a 3D-printed nano-strain sensor (5 x 2 µm²). The chamber was embedded into a circuit board chip and connected to an electronic read-out interface. The pressure signal was calibrated using a standardized gas-pressure protocol.
Result:
The collected signals showed a very sensitive and linear pressure to signal relation (r²=0.997), a high reproducibility (CV = 0.2 %), no hysteresis/drift over time, and a high stability even when performing continuous repetitive calibration procedures. The observed sensor specifications (pressure range: <-300 to >300 mbar, pressure resolution 480 µbar, membrane size 1x1mm²) would allow the sensor to track glucose changes with a resolution of 1-2 mg/dL.
Conclusion:
In conclusion, the first pilot attempt to miniaturize the core Sencell sensor technology by means of a nano-strain pressure sensor resulted in a very small osmotic pressure chamber (0.5 mm³) suitable for the anticipated purpose. Further miniaturization attempts are now underway to achieve the desired size specifications.
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
Pfützner Science & Health Institute Mainz, Germany
Background:
The TensorTip Combp glucose meter (CoG, CNOGA Medical, Cesarea, Israel) has been developed for invasive (INV) and non-invasive (NI) assessment of glucose in blood and tissue.
Method:
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
Institute for Diabetes Technology, Research and Development mbH at the University of Ulm
Ulm, Germany
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
Institute for Diabetes Technology, Research and Development mbH at the University of Ulm
Ulm, Germany
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
Institute for Diabetes Technology, Research and Development mbH at the University of Ulm
Ulm, Germany
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
Center for Diabetes Technology Charlottesville, Virginia USA
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
Department of Information Engineering, University of Padova
Padova, Italy
Objective:
The present work aims at improving prediction of hypoglycemia using continuous glucose monitoring (CGM) data only by exploiting individualized models and an ad-hoc identification cost function.
Method:
We considered a dataset composed by 177 CGM time-series collected in diabetic individuals with the Dexcom G6 CGM sensor for 10 days. For each individual, training set (32 hours) and test sets (8 days) have been created. For each individual, a personalized Auto Regressive Integrated Moving Average (ARIMA) model describing glucose dynamics has been identified on the training set. Model order has been chosen using 1-fold cross validation. Model parameters have been estimated by minimizing an ad-hoc cost function: the glucose mean square error (gMSE), that modifies the state-of-art cost function (Root Mean Square Error, RMSE) to account for the clinical impact of a prediction error. As a comparator, we considered a predictor that employs a state of art Auto Regressive (AR) model, whose parameters are identified by minimizing the RMSE. The models were applied in real-time settings on the test set to predict future glucose 30-minutes ahead in time. Hypoglycemic alarms have been triggered if predicted glucose fell below 70 mg/dl. Hypoglycemia prediction performances were compared using Precision, Recall, and F1-score.
Result:
The proposed ARIMA models outperform the comparator: in median [25th, 75th percentiles] Precision: 0.62 [0.48-0.75], Recall: 0.83 [0.73-1], and F1-score: 0.70 [0.59-0.78], whereas AR models: Precision: 0.41 [0.20-0.60], Recall: 0.28 [0.01-0.50], and F1-score: 0.40 [0.26-0.57].
Conclusion:
By using an ad-hoc cost function designed to account for the clinical impact of a prediction error, we can obtain more effective individualized models for predicting future hypoglycemic episodes.
Peripheral Focused Ultrasound Stimulation (pFUS): A New Therapy for Metabolic Dysfunction
General Electric Research Niskayuna, NY, USA
Objective:
A new study published in Nature Communications outlines our groups results using peripheral focused ultrasound stimulation (pFUS) within peripheral organs to precisely activate autonomic nerve circuits. We demonstrated the first study in which pFUS is utilized to stimulate a small sub-organ hepatic location containing sensory neurons. The ultrasound stimulus in this location was shown to modify the sensory input into the major brain centers controlling metabolic homeostasis and suppress the hyperglycemic effect of endotoxin exposure. Here we further address the capabilities of hepatic pFUS in the long-term management of metabolic disease.
Method:
Hepatic pFUS stimulation experiments were performed in the Zucker Diabetic Fatty rat model (ZDF rat; which develops significant obesity at 4 weeks of age and progressive development of hyperinsulinemia, hyperlipidemia, and impaired glucose tolerance). ZDF rats were treated with pFUS for 3 minutes a day targeted to the porta hepatis for 5.5 weeks. Animals were divided into two groups: a sham stimulated control and an early onset treated group. Circulating glucose, circulating insulin, and insulin-resistance scores (HOMA-IR) were measured.
Result:
Application of pFUS in the early onset treated group prevented the increase in circulating glucose (205 mg/dL at day 66 compared to 522 mg/dL in the sham control group). The pFUS early onset treated group showed significant improvements in insulin resistance as shown by a decrease of the HOMA-IR from a severe score of 2.9 (on Day 55) to a subclinical diabetic score of 1.5 (on Day 65). Furthermore, the hepatic pFUS stimulation resulted in an observable decrease in circulating insulin suggesting that glucose reduction was not the result of an increase in pancreatic insulin secretion.
Conclusion:
These results highlight the growing evidence that ultrasound energy, previously shown to enable activation or modulation of central nervous system pathways, may be used to perform peripheral neuromodulation in the treatment of metabolic disorders such as type 2 diabetes. The use of pFUS to directly modulate neuro-physiological systems therapeutically may provide alternatives to traditional pharmaceuticals. This nascent field will need to continue to develop a new understanding of how traditional drug pathways relate to the parameters, protocols, and outcomes of this new stimulation technology.
Glucose Clamp Quality Parameters among Study Populations
ProSciento, Inc. Chula Vista, CA, USA
Objective:
The euglycemic glucose clamp (GC) technique is the gold standard for the pharmacodynamic assessment of novel and biosimilar insulins. However, no currently accepted criteria exist to analyze clamp quality. Our objective was to assess clamp quality parameters among different subject populations from a significant pool of data and evaluate acceptability criteria.
Method:
We obtained coefficient of variation (CV) and deviation from target (DFT) from N=976 semi-automated (algorithm supported) euglycemic GCs using proprietary technology with the GlucoScout as the glucose sensor. These ≥24h clamps from 5 studies were performed to assess time-action profiles of long-acting insulins. CV is defined as the variability of blood glucose (BG) measurement: (Standard Deviation of BG/Mean BG)*100. DFT is defined as the difference between BG and target level: [Mean (Measured BG – target level)/target level]*100.
Result:
A total of 357 clamps in T1DM subjects, 177 in T2DM subjects, and 442 in healthy volunteers (HV) were analyzed. Overall mean CV was 5.37±2.11% (mean±SD); 95% CI (5.24, 5.51), and median was 4.98%. Overall mean DFT was 4.32±1.64% (mean±SD); 95% CI (4.21, 4.42) and median was 3.97%. Median CVs for T1DM subjects, T2DM subjects, and HV were 5.48%, 5.17%, and 4.32%, respectively. Median DFTs for the same populations were 5.8%, 4.26% and 3.46%, respectively.
Conclusion:
Clamp quality parameters assessing both the variability and the deviation from target should be analyzed in clamp studies. Minor differences were found among study populations; these are likely related to how the GC algorithm compensates for heterogeneity of the underlying subject phenotypes. We suggest that acceptability ranges for these parameters should be within 2 SD of the respective means (CV<10% and DFT<8%) and need not be modified for different subject populations.
The Role of Continuous Glucose Monitoring in Safety Management of Early Phase Clinical Trials
ProSciento, Inc. Chula Vista, CA, USA
Objective:
The use of continuous glucose monitoring (CGM) in clinical trials as an endpoint is well accepted. However, its role as a safety tool is underappreciated, particularly in early phase trials where pharmacodynamic parameters might be poorly understood or in dose ranging studies. We present preliminary data from a Phase 1, double-blind, clinical trial assessing escalating doses of a blood glucose (BG) lowering compound in subjects with type 1 diabetes (T1D) where a CGM protocol was utilized to support safety monitoring and prevent episodes of hypoglycemia.
Method:
Dexcom G4® PLATINUM CGM systems were used to monitor BG and guide preventive interventions. When BG levels <110 mg/dl and rate changes >1 mg/dl/min were detected, a preventive intervention (PrI) with rapid-acting oral carbohydrate was performed; nurses could also administer additional PrI if they felt subjects were at risk. Data from 12 subjects with T1D who completed 17 dosing periods are presented here.
Result:
Mean post-dose hypoglycemic events (HE) (plasma glucose (PG) concentration <70 mg/dl ± symptoms, or with symptoms and PG >70 mg/dl) was 4.29±2.97 (mean±SD) and mean nocturnal HE (midnight-6 am) of 0.83±1.13 per subject. The total average/day HE was 0.58±0.45. There were 391 PrI, averaging 23±9.34 per subject. The mean nocturnal PrI was 8.12±4.71 per subject. The daily average of PrI was 3.08±1.34. Subjects required 87.40±36.60 grams of additional carbohydrate/day to prevent HE. There were no episodes of severe hypoglycemia or serious adverse events (SAEs).
Conclusion:
If systematically utilized within standardized safety procedures, CGM can play a critical role as a safety tool and in furthering the understanding of pharmacodynamic time-action profiles in early phase trials with patients when developing new hypoglycemic agents.
Multi-National Performance Evaluation of the WaveForm Cascade (GlucoMen Day) CGM System
Agamatrix, Inc Wilsonville, Oregon, USA
Objective:
WaveForm is about to commercially launch the Cascade CGM I, trocar-free continuous glucose monitor (CGM) system for people with diabetes in Europe. The majority of the commercialization in Europe will be done by Menarini under the GlucoMen Day CGM product offering. Part of the process leading up to commercialization was to confirm safety and efficacy of the Cascade CGM system. A 14-day study was conducted in which a number of safety and efficacy parameters were assessed. One of these assessments was that of performance which was based on both in-clinic and home use data.
Method:
The multi-center study was conducted in clinical centers in Slovenia, Croatia, and Serbia (n=56 diabetic subjects). Each subject wore two Cascade CGMs in the abdomen over 14 days. Subjects used the system at home and in five in-clinic days (day 1, 4, 7, 10 and 14), which included 12 hours of blood glucose testing using a reference glucose analyzer. Accuracy assessment was based on YSI and CGM glucose pairs during in-clinic days and fingerstick BGM glucose and CGM glucose values for the home use part.
Result:
A total of 17,669 in-clinic CGM/YSI data points were analyzed. Mean absolute relative difference (MARD) and mean absolute difference (MAD) were 11.4% and 14.7 mg/dL. Consensus-error grid analysis showed 99.3% of data points were in zone A and B, 0.7% in C, in none in D and E. The system accuracy during home use was 12.7% and 15 mg/dL for MARD and MAD respectively, 98.6% of BGM/CGM readings were in zone A and B, 1.5% in zone C, and none in zones D and E.
Conclusion:
The performance of the Cascade CGM device over 14 days meets the safety and efficacy standards of CGM systems for managing blood glucose levels in people with diabetes. We expect to launch the Cascade CGM system in Europe in the third quarter of 2019.
Performance of the Bihormonal iLet Bionic Pancreas with the Stable Glugagon Analog Dasiglucagon
Massachusetts General Hospital Diabetes Research Center Boston, MA, USA
Objective:
We evaluated the function of the bihormonal iLet bionic pancreas delivering dasiglucagon when compared to the insulin-only iLet in a home-use study in adults with type 1 diabetes (T1D).
Method:
Ten subjects used the bihormonal and insulin-only configurations of the iLet for one week each in random order. The bihormonal iLet delivered 4 mg/dl dasiglucagon, a glucagon analog stable in aqueous solution (Zealand Pharma). Sessions were initiated by entering only the body weight; the iLet autonomously and continuously adapts to individual insulin needs. The primary outcome was iLet performance targets: continuous glucose monitor glucose (CGMG) readings capture ≥80%, drug delivery channel(s) availability ≥95%, and ratio of delivered to attempted doses 0.95 to 1.05. Secondary outcomes included percent of time <54 mg/dl, carbohydrates to treat hypoglycemia, mean CGMG, time in range (70-180 mg/dl), and percent of subjects with mean CGMG <154 mg/dl.
Result:
The iLet met the specified performance targets in both configurations. For the bihormonal vs. insulin-only configuration, median percent time with CGMG < 54 mg/dL was 0.2% [0,0.3] vs. 0.6% [0.2,1.1] (p=0.16); mean daily carbohydrate to treat hypoglycemia was 13.0±9.6 vs. 16.1±13.0 grams/day; mean CGMG was 139±11 vs. 149±13 mg/dL (p <0.01); mean percent time within the 70-180 mg/dl range was 79±9% vs. 71±8% (p<0.01); and percentage of subjects with mean CGMG <154 mg/dl was 90% vs. 50% (p=0.046). There were no infusion set occlusions or site reactions.
Conclusion:
The iLet performed to specifications in both the bihormonal and insulin-only configurations. Glycemic outcomes were very similar to those achieved with previous hardware implementations of the bionic pancreas. Dasiglucagon was well tolerated and the bihormonal iLet achieved similar glycemic outcomes to those previously achieved with freshly reconstituted human glucagon.
Use of the Ultra-Rapid Insulin Fiasp in the iLet Bionic Pancreas
Massachusetts General Hospital Diabetes Research Center Boston, MA, USA
Objective:
We evaluated the function of the insulin-only iLet bionic pancreas in adults with type 1 diabetes on pump or MDI therapy comparing the use of Fiasp in the iLet, and lispro or aspart in the iLet to their usual care and to each other in a home-use study.
Method:
This was a 3-way, random-order cross-over, home use study comparing the iLet delivering Fiasp (iLet-F) vs. the iLet delivering lispro (n=26) or aspart (n=8) per their usual insulin (iLet-LA) vs. usual care (UC) for 7 days each. Bionic pancreas sessions were initiated by entering only the body weight; the iLet autonomously and continuously adapted to individual insulin needs. The (pharmacokinetic) PK setting in the iLet algorithm was not adjusted for Fiasp.
Result:
The mean CGM glucose in the iLet-F arm (155±11, p=0.042), but not in the iLet-LA (155±13, p=0.097), was significantly lower than in the UC arm (162±26). There was no difference in mean CGM glucose between the iLet-F and iLet-LA arms (p=0.64). There were no differences in median % time <54 mg/dl between the arms (0.49 [0.0,1.0] vs. 0.53 [0.2,1.0] vs. 0.35 [0.1,1.2], p>0.64). The percentage time in the 70-180 mg/dl range was greater in the iLet-F (70.6±8.1%, p=0.001) and iLet-LA (70.1±9.2%, p=0.006) arms vs. the UC arm (61.5%), but there was no difference between the iLet-F and iLet-LA arms (p=0.54). There were no differences in mean insulin total daily dose (TDD) between arms (p>0.45).
Conclusion:
The iLet can provide effective glucose control when delivering Fiasp, insulin aspart, or insulin lispro. Adjustment to the PK settings of the iLet may be necessary to further improve glycemic outcomes when using Fiasp.
Predicting Mortality in both Diabetes Open-source Clinical Datasets from Free Text Entries using Machine Learning (natural language processing)
Diabetes Centre, Gartnavel General Hospital Glasgow, Scotland, UK
Objective:
We aimed to test the utility of semantic analysis to predict all-cause mortality from free-text entries from both a national diabetes database and an open source clinical dataset (MIMIC-III). We analyzed text entries alone in order to fully understand the potential of language analysis to predict outcome.
Method:
Diabetes dataset: An analysis period of 3 years was defined during which clinical text data were extracted. Mortality status at 1 year was identified. Data were preprocessed and divided randomly into training/validation and test sets 0.8:0.2. The training/validation set was further randomly divided 0.8:0.2. Dimensionality reduction was performed using embedding, and a combined convolutional and recurrent long short-term memory (LSTM) neural network was trained on the training subset for 20 epochs. Class imbalance was managed by applying class weights. A prediction of outcome was made on the withheld test set using the trained model, and area under receiver operator characteristic curve (AUROC) was calculated. For the MIMIC-III dataset, a similar methodology was applied, with some further development of sophistication of neural network architecture.
Result:
For the diabetes dataset: N=53,954 individuals with data were identified and N=2,292 deaths were recorded at 1-year post analysis. AUROC of model predictions when applied to withheld test set was 0.62. For MIMIC-III: 11,518 individuals were identified, with 2,045 deaths at 1 year. AUROC applied to withheld test set 0.86.
Conclusion:
By learning from clinician’s summaries, natural language processing (NLP) has the potential to leverage the clinical understanding of multiple clinicians and integrate information from multiple data sources. These models may be trained on outcomes such as mortality, aiding risk stratification, or on outcomes such as response to particular therapeutic agents, aiding clinical decision making.
Preventing Inpatient Hypoglycemia using Real-Time Continuous Glucose Monitoring: The Glucose Telemetry System
University of Maryland School of Medicine Baltimore, MD, USA
Objective:
Limited data on the utility of continuous glucose monitoring (CGM) in the inpatient setting is available. Our previously reported pilot study on the Glucose Telemetry System (GTS) demonstrated that real-time CGM data can be successfully transmitted wirelessly to the nursing station. We report the results from our ongoing clinical trial on the use of GTS in preventing hypoglycemia in patients with diabetes mellitus type 2 (DM2).
Method:
Insulin-treated patients with DM2 and high risk for inpatient hypoglycemia were randomized either to GTS or Point-of-Care (POC) testing. In the GTS arm, real-time CGM with alarms were set at 85 mg/dl alerting nursing staff to proceed with hypoglycemia preventative actions. In the POC arm, subjects used “blinded” CGM without hypoglycemia alarms. Hypoglycemic events (values <70 mg/dl for more than 15 minutes), mean percent time spent within target (70-179 mg/dl), and hyperglycemia (≥180 mg/dl) were evaluated.
Result:
A total of 37 patients with DM2 were randomized to GTS (n=18, intervention) and POC (n=19, standard of care). There were 34 preventative actions for impending hypoglycemia. The subjects in the GTS arm had 12 hypoglycemia events compared to 19 events in the standard of care arm (p=0.30). No difference between the GTS arm and the standard of care arm in time spent in normoglycemia or hyperglycemia were observed between the groups [mean(SD): 53.74% (21.47%) vs 50.49% (27.27%), p=0.71, and 46.00% (21.54%) vs 47.88% (27.98%), p=0.83, respectively].
Conclusion:
Preliminary analysis of our ongoing clinical trial showed that the use of real-time CGM (GTS) in the inpatient setting may lead to a reduction in hypoglycemic episodes without worsening overall glycemic control.
From Novice to Geek: Teaching Endocrinology Fellows How to Integrate Technology into Practice
New York-Presbyterian/ Weill Cornell Medicine New York, NY, USA
Objective:
The purpose of this pilot project is to evaluate a comprehensive educational intervention for endocrinology fellows to learn how to introduce technology into practice and then utilize patient-generated health data to inform shared decision making.
Method:
In July/August 2019, a weekly diabetes technology course with 60 minute sessions was given to two first- and two second-year endocrinology fellows at NewYork-Presbyterian/ Weill Cornell Medicine. A 20-question pre-test extrapolated from the Diabetes Technology Society Certification (CDTC) exam was given before the first session. A post-test will be given in September with the same questions using randomization of order of both questions and answer options. A brief survey measuring comfort with diabetes technology/pattern management will be given at the same time points. Participants are encouraged to wear devices. Descriptive statistics will be performed to examine differences between comfort and pre- and post-test scores with appropriate nonparametric statistical tests applied.
Result:
The median pre-test score was 77.5 (range: 65, 90). Second-year fellows scored higher on the pre-test compared to first-year fellows (scores 80 and 90 vs. 65 and 75, respectively). For the comfort with diabetes technology survey, the median (range) score was 6 (2, 10) for comfort with insulin pumps/delivery devices, 6 (4, 8) for blood glucose monitoring (BGM)/continuous glucose monitoring (CGM), and 3 (2, 8) for downloading devices/making treatment changes. Comfort level scores in all three categories were higher for second-year fellows compared to first-year fellows. Post-test data to follow September 10th.
Conclusion:
Diabetes technology has changed how we deliver care to people living with diabetes. Endocrinology fellows would benefit from knowledge of diabetes devices, patient selection, and data interpretation early in training. This curriculum can be adapted and utilized by other fellowship programs.
No Type 1 Left Behind: Implementing an Electronic Solution to Missed Basal Insulin Doses During Hospitalization
New York-Presbyterian / Weill Cornell Medicine New York, NY, USA
Objective:
Missed doses of basal insulin (MDBI) can lead to serious consequences such as ketosis. Our aim is to reduce the number of MDBI in medicine in-patients with type 1 diabetes (T1DM).
Method:
Two interventions were initiated to mitigate MDBI in patients with T1DM at New York-Presbyterian Hospitals (NYPH). The first intervention was an enhancement in the electronic health record (EHR) at all ten sites alerting the provider that their patient with T1DM should be placed on basal insulin. Medical logic memory (MLM) alerts “fired” if basal insulin was not ordered in a patient identified as having T1DM. The second intervention consisted of retrospective chart reviews at the Weill Cornell campus to see whether the alerts in the EHR were effective in reducing the number of MDBI. Reasons for missed doses were categorized as prescriber, pharmacy, nurse, or patient error and an incidence rate was calculated.
Result:
In the entire cohort (N=62), there were 14 episodes in 2017 and 5 episodes in 2018. The incidence rate of MDBI pre-intervention was calculated as 5 MDBI per 100 days vs. 3 MDBI per 100 days post-intervention (p=0.152). Of the 14 episodes in 2017, N=6 were due to prescriber error, N=1 to nurse error and N=7 to patient refusal of insulin (same patient). Of the 5 episodes in 2018, N=4 were due to prescriber error and N=1 to pharmacy error.
Conclusion:
The most common reason for MDBI was prescriber error, regardless of year. There is a trend towards decreased incidence post EHR interventions. Our results underestimate the number of MDBI due to our small cohort and possible ICD-10 miscoding of patients with T1DM. Our next step is to review surgical patient EHRs.
Incorporating the Effects of Psychological Stress Response on Glucose Predictions
Illinois Institute of Technology Chicago, IL, USA
Objective:
Acute psychological stress (APS) induces metabolic responses that increase the risk of glucose dysregulation in people with diabetes. APS is difficult to detect non-invasively and its effect on glucose concentration (GC) dynamics is not well quantified. An APS detection and assessment algorithm was developed to quantify APS levels from physiological measurements from a wearable activity tracker (Empatica E4). The APS assessments were used as inputs to a GC prediction model to quantify the effects of the APS on the GC.
Method:
Experiments were conducted with standard APS inducing techniques while collecting physiological, continuous glucose monitor (CGM) measurements, and insulin information. A wristband was used to collect biosignals. Over 800 features are extracted from pre-processed biosignals. Machine-learning-based APS detection and evaluation algorithms were developed to quantify the effects of APS with an APS effect index (APSEI) and to capture its effects on GC predictions. We incorporated APSEI values as an additional input to our GC prediction model.
Result:
APSEI is statistically significantly different during APS inducement periods and the rest of the experiment (p-value < 0.05) and it is used as an input for CGM prediction. Support vector machines yields the best APS detection results and Bayesian optimization is utilized to achieve high accuracy. The use of APSEI improves the CGM predictions by more than 2% on average.
Conclusion:
The APS assessments are used as inputs to a GC prediction model to quantify the effects of the APS on the GC and improve CG prediction.
Incorporating Wearable Physical Activity Trackers to Improve Glucose Predictions for Multivariable Artificial Pancreas Systems
Illinois Institute of Technology Chicago, IL, USA
Objective:
The type, intensity, and duration of physical activities affect blood glucose concentration (BGC) of people with type 1 diabetes (T1D). An accurate estimation of exercise type and intensity is necessary for improving the accuracy of BGC estimation and the performance of artificial pancreas (AP) systems without exercise announcement. Signals from a non-invasive wristband is used to classify physical state (PS), estimate energy expenditure (EE), and estimate BGC.
Method:
Twenty-two experiments with aerobic and resistance exercise sessions are conducted with 7 different subjects with T1D. Empatica E4 wristbands were utilized to collect biosignals as streaming data. Over 800 features were extracted from five different denoised signals (skin temperature, accelerometer, galvanic skin response, blood volume pulse, heart rate). A machine-learning-based PS classification algorithm was developed for distinguishing five different activities with the extracted features. The EE estimation algorithm was developed using similar features. Various machine learning techniques were used. The PS and EE estimates were used in BGC prediction for use in multivariable AP systems.
Result:
The PS and EE estimates used as inputs to a BGC prediction model resulted in an average improvement of over 15% in mean absolute percentage error in glucose concentration prediction accuracy.
Conclusion:
Biosignals of wristbands can be used to determine exercise type and intensity. These estimates provide advance information for better BGC predictions and insulin dosing decisions for AP systems without announcements of exercise to offset the delays between the occurrence of activities and the BGC and CGM information.
Mobile-enabled Food Logging is Associated with Improved Glycemic Management in the Real-world
Glooko, Inc. Mountain View, CA, USA
Objective:
Understanding the relationship between diet and glycemia is an important part of diabetes self-management. In the current study, we assess whether the use of an in-app food logging feature corresponded with improved glycemic outcomes.
Method:
We randomly sampled 1,000 adult users of a diabetes management mobile app and investigated whether in-app food logging was correlated with self-monitoring of blood glucose (SMBG) behavior and glycemic outcomes [i.e., mean blood glucose, the proportion of hypoglycemic (<70 mg/dl), in-range (between 70-180 mg/dl), and hyperglycemic (>250 mg/dl) SMBG readings] during the first 90 days of mobile app use.
Result:
Among those with demographic information, 45.2% were female, median age was 47 years (IQR: 35-57 years), and 56.4% self-reported to have type 2 diabetes (27.1% with type 1 diabetes, 16.5% with other diabetes). Food logging was correlated with SMBG check frequency (Spearman’s ⍴ = .16), the proportion of in-range SMBG readings (⍴ = .11), mean blood glucose (⍴ = -.13), and the proportion of hyperglycemic readings (⍴ = -.11; all P values <.001). Accounting for SMBG check frequency, food logging was still associated with in-range readings (T = 3.48), mean blood glucose (T = -3.46), and hyperglycemic readings (T = -3.38; all P values<.001). Quartile comparisons between frequent and infrequent food loggers showed that frequent loggers performed SMBG checks more frequently (median 3.0 vs. 2.0 checks/day, P<.001) and had lower mean blood glucose (median 143.2 vs. 137.2 mg/dl, P = .047).
Conclusion:
Our results suggest that mobile-enabled food logging can be a valuable part of glycemic management support in the real world. As mobile health app features and capabilities continue to evolve, they have the potential to encourage greater engagement in self-management that can lead to substantive improvements in diabetes management.
A Novel Method for Determination of Pharmacodynamic Onset of Prandial Insulin Action
Profil Institute of Metabolism Research Neuss, Germany
Objective:
Recently, ‘Onset of Insulin Action’ in the glucose Clamp setting was arbitrarily defined by a decrease from baseline (DFB) blood glucose (BG) concentration of 5 mg/dL. A limitation of the current definition is that Onset of Action may be under- or overestimated when baseline BG levels show either increasing or decreasing trends. Here we introduce a novel method, based on a change in trend (CIT), for determination of pharmacodynamic onset of insulin action and its validation in in vitro Clamp Experiments.
Method:
The CIT method, programmed in R, minutely computes a 95% prediction interval based on linear regression of the preceding 30 BG values. After dosing, the first time point a measured BG value falls below the prediction interval is considered pharmacodynamic onset of insulin action. To validate this method, we conducted 18 in vitro clamp experiments at different rates of change in BG. The obtained clamp data were analyzed for time of onset with both methods and then further stratified for decreasing, stable, and increasing BG trends.
Result:
Median onset was comparable for CIT (28.0 (IQR 2.75) min) and DFB (29.0 (IQR 34.00) min). However, mean deviation from median was much larger for DFB (14.83 vs. 1.78 min) and there were large differences in onset if pre-dosing BG decreased [median onset 28.0 (CIT) vs. 4.5 min (DFB)] or increased (27.0 vs. 41.0 min).
Conclusion:
CIT provides a more robust determination of onset of action than DFB, by incorporating baseline BG changes in its calculation. This novel method may be used in real time for both automated and manual clamp studies to strengthen clinical pharmacology methodology and improve pharmacodynamic profiling of insulin products.
Accuracy of the CONTOUR®NEXT ONE Blood Glucose Monitoring System in the Low Blood Glucose Range Using Probability Methodology
Ascensia Diabetes Care US Inc., Global Medical Affairs Parsippany, NJ, USA
Objective:
Data analyzing blood glucose monitoring system (BGMS) accuracy in the low blood glucose range (LBGR: ≤70 mg/dL) are lacking, warranting further research. Not all BGMSs meeting FDA 2016 and/or ISO 15197:2013 accuracy criteria yield reliably accurate results in the LBGR. This post hoc analysis utilizes previously presented probability methodology to estimate the likelihood of accurate BGMS performance in the LBGR.
Method:
Data were computed from fingertip capillary blood samples obtained by study staff from diabetes patients in two separate trials. Trial 1 (Christiansen M, et al. J Diabetes Sci Technol. 2017;11:567-573) was conducted in the US and included the CONTOUR®NEXT ONE (CNO) BGMS only. Trial 2 (Jendrike N, et al. Curr Med Res Opin. 2019;35:301-311), conducted in Europe, compared five systems including the CNO BGMS (and is included here to corroborate results from Trial 1). To estimate likelihood of accurate BGMS performance (results ±15% of reference values) across an entire range of BG concentrations (20–460 mg/dL), probability curves were computed based on a linear regression model with BGMS results expressed as a function of laboratory data.
Result:
For the CNO BGMS, the probability of accurate system performance at specific BG concentrations in the LBGR (40, 54 and 70 mg/dL) was >95% in both trials.
Conclusion:
In this analysis, the CNO BGMS was highly accurate in the LBGR, which is important for safe and effective diabetes management – especially in insulin-treated patients, diabetes patients with history of severe hypoglycemia or hypoglycemia unawareness, diabetes during pregnancy, and/or patients using continuous glucose monitor (CGM) when BGM monitoring is recommended. This analysis also further demonstrates the utility of probability methodology for assessing BGMS accuracy.
Past, Present, and Future of Diabetes Technology in Developing Countries
National University of Medical Sciences Islamabad, Punjab, Pakistan
Introduction:
Diabetes Mellitus (DM) is a global public health concern with a rising prevalence and long lasting impairments. In the past two decades, there has been tremendous progress in the medical sciences and diabetes technology (DT) in particular. The modern techniques of diagnosis and management of DM are flourishing in the developed countries but the developing nations are still lacking the benefits of DT in the management of DM. This study’s aim was to investigate the knowledge, attitude, and practice of DT amongst medical students in the developing world.
Method:
A survey was conducted at the National University of Medical Sciences, Rawalpindi, Pakistan, amongst medical students from 1st to 5th year. Students were asked opened ended and short phrase questions on DT, diagnostic tests, modern line treatment, the use of insulin pumps, and other modern DT techniques.
Result:
Responses received from 132 medical students showed that 90% of the students were aware of the alarming prevalence of DM in the developing world and its diagnostic tests. Seventy percent of students provided satisfactory answers for the modern line treatment of DM, 94% of students were aware about the usage and benefits of modern insulin pumps, and 100% acknowledged the beneficial use of smart insulin pens. However topics such as artificial pancreas systems and augmented pump therapy were less known with only 40% of students having knowledge of their implementation.
Conclusion:
The overall results indicated that medical students during their study of medicine have sufficient knowledge for “Diabetes Technology”. Developing nations must be further made a specific target for the implementation of DT and its beneficial uses in medical school curricula and clinical settings. Whilst considering the past and looking into the future, we have to be careful not to simply reflect our thoughts on what is left behind, but to consider the challenges and the activities we have to initiate to be able to handle this epidemic safely. In future, DT will be more than beneficial to not just the developing world but also to the far-fetched marginalized nations.
Painless, Bloodless Noninvasive Glucose Monitoring System Using a Fluorescent Glucose Binding Protein
University of Maryland Baltimore County Baltimore, MD, USA
Objective:
The overall objective of this project is a painless, bloodless, and safe noninvasive glucose monitoring system for the more delicate patients in our healthcare system: children, the elderly, and neonates.
Method:
A fluorescent glucose binding protein (FGBP) was used to develop a method for measuring transdermal glucose (TG) that has been collected by passive diffusion through the skin. A 250 µL volume of buffer in a 2.5-mL tube was allowed to come in contact with skin, collected, and analyzed with the FGBP. Correlation between TG and blood glucose (BG) was determined during Oral Glucose Tolerance Tests in adult human subjects, as well as, during standard glucose monitoring of patients in the pediatric intensive care unit. Experiments on a skin model using a flow cell was done to corroborate the human subjects testing. The development of the glucose sensing device using the FGBP will be described.
Result:
The concentration of TG was found to be in the micromolar range, three to four magnitudes less than BG. The TG concentrations are all within the linear range for the sensitivity of the fluorescent glucose binding protein (Kd ~ 1 µM). Nevertheless, correlation between TG and BG was found to be linear (R~1.0). Differences in skin diffusivity also determines the concentration of TG collected. However, skin diffusivity can be measured independently from the Transepidermal Water Loss (TEWL). In in vitro flow cell experiments using pig skin as a model, the ratio of TG/BG was found to be constant at a constant TEWL. Thus, it is possible to use TEWL as a reference in determining BG from TG.
Conclusion:
We have found, from multiple human subject experiments, that TG and BG are linearly correlated. With adjustments to skin diffusivity, it is possible to follow changes in BG by noninvasively measuring TG. An automated device for TG determination is being developed for bedside noninvasive glucose determination in the pediatric and neonatal intensive care units.
Glycated Protein Biosensing based on Engineering Enzymes for 2.5th Generation Amperometric Electrochemical Principle
Tokyo University of Agriculture and Technology Koganei, Tokyo, Japan
Objective:
Glycated proteins, such as glycated albumin (GA) and hemoglobin A1c (A1c), are important glycemic control markers for diabetes mellitus. Recently enzymatic assays have been introduced, which employ fructosyl amino acid/peptide oxidases. This study is focusing on the development of a novel and versatile electrochemical sensing principle for glycated proteins based on enzymatic methods, using engineered enzymes combined with a new redox probe, making enzymes capable of quasi-direct electron transfer (i.e.,2.5th generation principle).
Method:
Based on the 3D structures and their mutational simulations, we designed the positions where: 1) repressing the enzyme activity using oxygen as the electron acceptor but remaining its activity as dehydrogenase and 2) novel redox probe should be modified. As the new redox probe, phenazine ethosulfate; 1-[3 Succinimidyloxycarbonyl) propoxy]-5-ethylphenazinium triflate (amine reactive PES: arPES), was used.
Result:
The designed fructosyl peptide oxidase (FPOx) was recombinantly produced as the engineered “dehydrogenase” which showed quasi-direct electron transfer with electrode. Thus prepared enzymes were immobilized on the gold electrode using self-assembled monolayer (SAM) equipped in the flow cell and a flow injection analysis (FIA) was performed. A good linear range of 50 – 500 µM for fructosyl valine, the protease digested derivative of A1c, was achieved. The FIA system can be repeatedly used for more than 200 consecutive sample analyses.
Conclusion:
The design concept of engineered fructosyl amino acid/peptide oxidases realized the creation of biosensing molecules suitable for 2.5th generation electrochemical sensing principle of glycated protein analyses.
Risk-based Dosing of Insulin and Nutrition Improves Glycaemic Control Outcomes
University of Canterbury, Mechanical Department Christchurch, New Zealand
Objective:
Hyperglycemia and insulin resistance are common in critically ill patients and associated with worsened outcomes. STAR (Stochastic TARgeted) glycaemic control (GC) has proven effective over different units and clinical practices. Unlike many protocols, STAR also modulates nutrition with insulin, using a patient-specific risk-based dosing approach to provide greater flexibility in control. This study compares and assesses safety and efficacy of the ongoing STAR clinical trial results at the University Hospital of Liège, Belgium.
Method:
Two arms are compared: the first used an insulin only version of STAR (STAR-IO), and the second the full insulin+nutrition version of STAR. The target glucose band was 80-145mg/dL. Insulin was administered via IV and nutrition was administered enterally. GC was stopped after 72h or if BG was stable at insulin rate ≤2U/h. Safety was assessed by %BG <80mg/dL below target and hyperglycaemia (%BG>180mg/dL). Performance was evaluated by %BG within target band and median BG. Clinical data from 11 patients on STAR-IO and 10 patients on STAR totalling 1100 hours of control were used. Ethics approval was granted by the University Hospital of Liège Ethics Committee.
Result:
STAR performance is statistically significantly better compared to STAR-IO (89% vs. 78% for %BG in target band, p<0.01 using Fisher Exact test). Median [IQR] BG is similar but tighter in STAR (118[109 129] vs. 120[107 138]mg/dL, p=0.19 using Wilcoxon rank sum test). STAR is also safer compared to STAR-IO with 0.7% vs. 1.4% for %BG<80 mg/dL and only 2.0% vs. 9.8% for %BG>180mg/dL. This outcome was achieved using less insulin and nutrition rates for STAR vs STAR-IO (3.0[2.0 4.0] U/h vs. 3.5[1.5 6]U/h and 7.0[4.7 8.2] vs. 8.1[4.9 9.2]g/h).
Conclusion:
Modulating nutrition in addition to insulin can significantly improve GC outcomes, especially by reducing nutrition rates for highly resistive patients.
Women have Greater (Metabolic) Stress Response than Men
University of Canterbury, Mechanical Department Christchurch, New Zealand
Objective:
Stress hyperglycaemia is frequent in intensive care unit (ICU) patients and associated with increased morbidity and mortality. Glycemic control (GC) has proven difficult due to high levels of inter- and intra- patient variability in response to insulin. However, despite anecdotes, no one has studied if males or females are easier/harder to control. This study examines differences in clinically validated insulin sensitivity (SI) and its variability between males and females as surrogates of control difficulty.
Method:
Data from N=145 SPRINT GC patients were analysed for the first 72hours of stay. Demographic characteristics of the male (N=91) and female (N=54) sub-cohorts are similar (age, mortality, injury severity, ICU length of stay, GC duration), as well as GC outcomes (median BG, %BG in/out target band, workload). SI was identified hourly and its hour-to-hour percentage variability was computed (%ΔSI). Due to large data samples, the 95%CI of difference in bootstrapped medians in SI and %ΔSI was used for hypothesis testing to a significance level of p<0.05. Equivalence testing was used to determine whether this difference is clinically significant.
Result:
Females are more insulin resistant (lower SI) than males (2.5e-4[1.5e-4 4.0e-4] vs. 3.1 e-4[1.7e-4 5.5e-4] L/mU/min). This difference is statistically different and clinically not equivalent. Conversely, %ΔSI is not significantly different (2[-17 22]% vs. 3[-14 25]%), and any difference can be considered clinically equivalent. These observations are also true when data are analysed over 6-h blocks.
Conclusion:
Females are more insulin resistant than males but have equivalent SI variability. The difference in SI levels suggests either higher endogenous glucose production and/or lower insulin secretion rates for females. Since severity of injury and glycemic outcomes are similar across both groups, the results suggest a stronger stress response to injury for female patients.
Durable Glycemic and Hepatic Improvements After Duodenal Mucosal Resurfacing in Patients with Type 2 Diabetes
Gastroenterology and Hepatology, Amsterdam University Medical Centers, Academic Medical Center
Amsterdam, The Netherlands
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
ARTORG Research Center for Biomedical Engineering, University of Bern
Bern, Switzerland
Objective:
To compare the estimates of goFOOD with the visual estimates of experienced dietitians in terms of macronutrient (carbohydrate, protein, fat) and energy content.
Method:
The goFOOD application (an upgraded version of GoCARB) is an artificial-intelligence (AI)-based system running on Android smartphones and is designed to estimate macronutrient and energy content from food images: each plated food item is detected, recognized, segmented, and the respective volume estimated. In this study, the goFOOD application was employed to automatically estimate the volumes of 146 unique meal images, containing 1 to 6 different food items each, captured under free-living conditions in Switzerland. Each food item was segmented and recognized with integrated semi-automatic functionalities. Finally, this information combined with data from the USDA Nutrient Database was used to estimate the nutrient content of each food item. Two experienced dietitians (one from Switzerland and one from the USA) visually estimated the macronutrient and energy content of the same food items as the comparator.
Result:
The median difference (25th quantile, 75th quantile) between the estimates of goFOOD and of the dietitians, and the p values in a paired t-test are as follows: carbohydrates (g): 14.4 (4.8, 25.3) p=0.17, protein (g): 5 (1.8, 9.4) p=0.44, fat (g): 5.1 (1.3, 10.8) p=0.41, energy (kcal): 115 (41, 220) p=0.43. Overall the two methods do not differ significantly in macronutrient and energy estimations.
Conclusion:
This preliminary investigation indicated that goFOOD and dietitians provide similar estimates of nutrient content of foods (no statistical differences), implying that goFOOD could be a useful tool for dietary monitoring and assessment. Further research of the system’s algorithmic optimization and extension is ongoing, and is accompanied by a more extensive evaluation.
Enhancement of the Type 1 Diabetes Patient Decision Simulator to Describe the Behavior of Patients on Multiple Daily Injections
Department of Information Engineering, University of Padova Padova, PD, Italy
Objective:
The type 1 diabetes patient decision simulator (T1DPDS) can be used to develop/test insulin therapies on realistic multiple-day scenarios. The physiological model allows simulating both insulin pump and multiple daily injections (MDI) users, thanks to the recent incorporation of a long-acting insulin pharmacokinetic model (Visentin et al., TBME, 2019). The patient’s behavior and treatment decision model, however, allows computing insulin and carbohydrate doses in a reliable way for insulin pump users only (Vettoretti et al., TBME, 2018). Our aim is to equip the T1DPDS with a behavioral model of MDI patients to enable realistic in silico trials with MDI.
Method:
From the databases of the T1D Exchange Registry (2010-2012) and the T1D Management questionnaire (2015), we extracted information related to the behavior of MDI patients in making treatment decisions that was used to modify the existing behavioral model. Specifically, we updated meal insulin boluses, correction boluses, and hypoglycemia symptoms models, and introduced new models of missing meal boluses and missing basal injections. Results of a one-week simulation in 100 virtual adults were compared to the glycemic outcomes observed in the DIAMOND study (MDI+CGM group).
Result:
As expected, the glycemic outcomes obtained by the pump behavior model were significantly different from those reported in the DIAMOND study (time spent in eu-, hypo-, hyper-glycemia: 928 vs. 736, 23 vs. 43 and 477 vs. 632 min/day), whereas those achieved with the new MDI behavior model were much more similar (815, 23 and 574 min/day, respectively).
Conclusion:
We have enhanced the T1DPDS by including a new model of the MDI patient behavior. The T1DPDS can now be used to perform realistic in silico trials on MDI patients.
Modeling the Dexcom G6 CGM Sensor Error
Department of Information Engineering, University of Padova Padova, PD, Italy
Objective:
When compared with frequent blood glucose concentration references, continuous glucose monitor (CGM) measurements are unavoidably affected by error. The availability of CGM error models can be important in several applications, e.g. for testing in simulation the safety and effectiveness of CGM-based applications (e.g., decision support systems). In this work, we model the error of the Dexcom G6 sensor, which does not require in vivo calibrations and lasts for 10 days.
Method:
The dataset consisted of CGM and YSI values collected in parallel during 12-h clinical sessions in days 1 (or 2), 4 and 10 of the monitoring period in N=79 individuals with diabetes. The sensor error model was derived by using a methodology previously developed by our research group (Facchinetti et al., TBME, 2014), which considers 3 main error sources: the effect of plasma-interstitium kinetics, modeled by a first-order linear dynamic system; the calibration error, modeled by two time-varying polynomials of order m and l; and the random noise, modeled by an autoregressive model of order q. In this work, the calibration error model was extended to cover the entire sensor lifetime, to handle the data of calibration-free sensors.
Result:
Results of model identification show that the time-variability of sensor calibration error during the sensor lifetime can be well represented by a model with constant intercept (l=0) and slope described by a 2nd-order polynomial (m=2). The sensor noise is well described by an autoregressive model of order q=2.
Conclusion:
We derived a model of the Dexcom G6 sensor error that can be embedded into type 1 diabetes (T1D) simulators to enhance the reliability of in silico trials.
In Silico Optimization of Long-Acting Insulin Injection Time in Subjects With Type 1 Diabetes
Department of Information Engineering, University of Padova Padova, Padova, Italy
Objective:
Long-acting insulin glargine U100 (Gla-100) is commonly employed in multiple daily injection (MDI) therapy to cover basal insulin needs in subjects with type 1 diabetes (T1D). Although most T1D subjects inject Gla-100 in the evening/bedtime, some people could benefit from a morning injection depending on the diurnal variability of insulin requirements. We compared in silico the effect of morning vs. bedtime injection of Gla-100 on daily glucose control, in light of the circadian changes in insulin sensitivity (SI).
Method:
One hundred in silico T1D adults, each characterized by a specific SI diurnal pattern, underwent a two-arm 2-week 3-meal/day trial under MDI with the UVA/Padova Simulator (Visentin et al., JDST 2018). Fast-acting insulin Lispro was used for insulin meal boluses, while Gla-100 was injected either at bedtime (9.30-11.00pm) or in the morning (6.30-8.00am). The optimal Gla-100 injection time was selected for each subject as the one minimizing percent time of glucose below 70 mg/dL (Tb) and maximizing percent time of glucose in target [70-180] mg/dL (Tt). The distribution of optimal injections was stratified based on the diurnal pattern of subject’s SI.
Result:
Glucose control over the last week significantly improved in 39 subjects using Gla-100 in the morning, with average Tb reduced by 47% and average Tt increased by 12% compared to bedtime injection. Of note, subjects exhibiting low SI at breakfast and lunch compared to dinner were those mostly taking advantage of the morning injection.
Conclusion:
Morning injection of Gla-100 is effective in T1D subjects with low insulin requirements in the morning and early afternoon. These results can be translated in real life to optimize Gla-100 injection time based on subject’s SI variability pattern.
Accuracy of Bolus Delivery of a Novel Patch Pump for Insulin Delivery
Institute for Diabetes Technology, Research and Development mbH at the University of Ulm
Ulm, Germany
Objective:
An accurate delivery of insulin is important for an effective therapy and safety of patients using an insulin pump. A novel patch pump was recently introduced and tested for accuracy.
Method:
Delivery accuracy of the Accu-Chek® Solo micropump was tested in an experiment based on IEC 60601-2-24. Several boluses of different sizes (0.2 U, 1 U and 10 U) were delivered by three exemplars of the pump and weighed individually. In total, 225 deliveries were evaluated for the smaller boluses and 108 deliveries for the 10 U bolus. The total mean deviation and percentages of boluses within ±15% of target were calculated. This range was considered appropriate as IEC 60601-2-24 does not describe any acceptance criteria. In addition, time to deliver a 10 U bolus was recorded.
Result:
Mean deviation from target delivery was -3.3% for the 0.2 U bolus, 0.3% for 1 U and 0% for 10 U. Regarding the respective target, 88% of 0.2 U boluses, 99% of 1 U boluses, and 100% of 10 U boluses were within ±15%. Delivery time of a 10 U bolus was 4 min.
Conclusion:
In this evaluation, the novel patch pump accurately delivered the programmed boluses. No critical deviations were observed. Larger boluses were delivered more accurately, on a percentage base, than smaller boluses. This was already observed previously for other insulin pumps, and should be considered when using small doses in clinical practice.
Blood Glucose Prediction from Pooled Continuous Glucose Monitor Data
Technion
New York, NY, USA
Objective:
Proactive self-care prevents hypo- and hyperglycemia, improving time-in-range and A1c. Blood glucose monitors (BGMs) and continuous glucose monitors (CGMs) display current blood glucose (BG) levels, making self-care reactive (treating lows/highs) not proactive (preventing lows/highs). Without look-ahead BG data, people act on inaccurate guesses. We used app-entered data and BG readings from CGMs to more accurately predict near-term BG levels.
Method:
The One Drop app uses Apple’s Healthkit and Dexcom’s Application Programming Interface (API) to passively read BGs from CGMs. Data included contextual information, self-care, and BG values from over 4,000 One Drop users with CGMs and an individual whose BGs were to be predicted. Data from a random selection of 75% of sampled users trained a supervised learning model. The trained model with subsequent inputs from those same “seen” users generated predicted or forecasted BG values 30 and 60 minute into the future. BG values for “unseen” users, whose data were not used to train the model, were also forecasted.
Result:
Users were 86% type 1/LADA, 8% type 2, and 6% unreported; 28% were diagnosed ≤5 years. The mean absolute relative difference (MARD) of 60-minute predictions relative to CGM-reported values for seen (unseen) users was 14.2% (14.4%), with 77.6% (76.9%) of predictions falling in Zone A of the Clarke Error Grid, and 98.6% (98.4%) in Zone A or B. The MARD for 30-minute predictions was 5.5% (5.7%), with 96.8% (96.7%) of these in Zone A, and 99.8% (99.8%) in Zone A or B.
Conclusion:
Pooling BG data from thousands of One Drop app and CGM users confers accurate, short-term BG forecasts. These forecasts can inform proactive, preventive self-care.
Developing a Branched-Chain Amino Acid Biosensor with Bacterial Periplasmic Binding Proteins for Predicting Patient Risk for Prediabetes and Type 2 Diabetes
Center for Advanced Sensor Technology, University of Maryland Baltimore County
Baltimore, MD, USA
Objective:
The development of complications in diabetic patients is highly reliant on their prognosis. Altered levels of branched-chain amino acids (BCAAs) have been indicated as biomarkers for the onset of type 2 diabetes and its progression. However, current detection and quantification methods have long turnaround times and require expensive laboratory instrumentation and skilled operators. As such, there is a huge need for the development of minimally invasive, real-time detection and quantification tools for biomarkers such as BCAAs in point-of-care medical devices that are accessible to the public.
Method:
In this study, we are developing a BCAA biosensor derived from the leucine/isoleucine/valine-binding protein (LivJ) of Escherichia coli. Cysteine mutations are strategically placed close to the binding pocket of LivJ and these sites are subsequently labeled with a polarity-sensitive dye. Mutants are evaluated for their detection range, sensitivity, and stability. Select mutants will be implemented in an immobilized protein platform for integration into a fully functional, stand-alone sensor as a plug-and-play module.
Result:
The mutant LivJ candidate shows an increasing fluorescence response with increasing BCAA concentration in the micromolar range, exhibiting a linear range of 0 – 100 uM. Preliminary studies for the development of a protein immobilization platform showed an increased detection range of 0.01 – 2 mM and a 40% improvement in the response amplitude.
Conclusion:
This highlights the potential of LivJ as a regenerable biosensor for implementation in medical devices utilized in diabetes preventive care. Further studies are in progress to characterize the protein and evaluate its potential in a clinical context.
“Smart” Composite Microneedle Patch Delivers Thermostable Glucagon for the Prevention of Nocturnal Hypoglycemia
Advanced Pharmaceutics & Drug Delivery Laboratory, University of Toronto
Toronto, Ontario, Canada
Objective:
Hypoglycemia is a major complication associated with insulin therapy in diabetes. It could cause life-threatening conditions in patients and significant anxiety and fears to both patients and caregivers. Despite the availability of glucagon kits for severe hypoglycemia rescue, the inability to administer glucagon, hypoglycemia unawareness, or hypoglycemia during sleep put the patients at high risk. To prevent episodes of unattended hypoglycemia, we designed a “smart” composite microneedle (MN) patch capable of stabilizing glucagon, sensing hypoglycemia automatically, and delivering glucagon in response to low glucose.
Method:
Native glucagon was encapsulated in polymeric microgels containing a glucagon-stabilizing component. The composite MN patch was prepared by thermal cross-linking of poly(methylvinylether-co-maleic anhydride) with poly(ethylene glycol) with incorporation of glucagon-loaded glucose-responsive microgels. Stability of released glucagon was determined by HPLC and far-UV circular dichroism (CD) spectropolarimetry. The properties of composite MN arrays were fully characterized in vitro. The glucose-responsiveness and hypoglycemia prevention in vivo were evaluated in an STZ-induced type 1 diabetic rat model.
Result:
Glucose-responsive microgels were successfully synthesized with a relatively uniform size and incorporated into the MN patch. Composite MNs were mechanically strong and able to penetrate shaven rat skin. Glucagon release from the composite patch was more rapid at hypoglycemia than at high glucose levels. Released glucagon from the composite MN patch maintained structural stability compared to the control. The composite MN patch could prevent insulin-induced hypoglycemia for multiple hours after patch application.
Conclusion:
A “smart” composite MN patch embedded with glucose-responsive and glucagon-stabilizing microgels has been successfully designed for the prevention of hypoglycemia for a duration of overnight sleep. This system is promising to reduce the risk of hypoglycemia in patients with diabetes and the burden of their caregivers.
Clinical Outcomes of V-Go® Wearable Insulin Delivery Device Based on Baseline TDD in Patients with Type 2 Diabetes
Premier Physician Network, Bull Family Diabetes Center Dayton, OH, USA
Objective:
Determining insulin requirements can be challenging. Prescribed doses are often significantly higher than actual doses administered, especially in higher doses. This non-adherence leads to false perceptions of insulin needs and potential overinsulinization in an effort to improve glycemic control. Use of V-Go, a wearable insulin delivery device has resulted in improved adherence and lower insulin requirements. This evaluation investigated the relationship between baseline insulin total daily dose (TDD) and change in A1C and TDD when V-Go is used for insulin delivery.
Method:
Electronic medical records from a type 2 diabetes population prescribed insulin were evaluated. Patients not controlled (A1C >7%) and switched from basal-only, basal-bolus, or premix insulin regimens to V-Go were included. Patients were stratified according to baseline prescribed TDD (< 50 U/day, 50 to 100 U/day or > 100 U/day). Paired t-tests were used to evaluate clinical outcomes including change in A1C and TDD by baseline TDD.
Result:
Patients (N=122) were evaluated after a mean duration of 5 months. Percent of patients in each baseline TDD stratum was < 50U=41%, 50 to 100U=45% and > 100U=14%. Significant reductions in A1C were observed across all strata and a significant reduction in TDD was reported for strata 50 U or greater on V-Go. Patients prescribed >100 U at baseline benefitted from an A1C reduction of 1.3%; p<0.05 utilizing 47% less insulin (135 to 72 U/day; p<0.0001).
Conclusion:
V-Go improved control across all TDD strata and patients with higher baseline doses benefited from clinically meaningful glycemic improvements with significantly less insulin.
Clinical Studies of Extended Wear Adhesive Patches for Use on Insulin Infusion Set
Medtronic Diabetes Northridge, CA, USA
Objective:
Extending insulin infusion set (IIS) wear time from 2-3 days to 7 days to match the wear time of glucose sensor is a desirable feature for insulin pump users, especially for sensor-augmented pump (SAP) users. Longer wear time means less injuries of the skin, less hassle for device change, and lower costs per day. However, longer IIS wear time means also higher challenges for the adhesive material used. During wear, the bonds between the adhesive patches and skin are constantly challenged by external factors which cause the edges of the adhesive patch to lift off of the skin, making the device vulnerable to premature fall off. In addition, the consequence of longer usage time might be more skin reactions (e.g. irritation, sensitization, etc.) and less comfort. The objective of this study was to evaluate baseline performance of the IIS adhesive patches currently on-market and to test the on-body performance of the new IIS adhesive patches for extended wear up to 8 days.
Method:
A total of 75 adults were recruited to wear the various adhesive patches on several Medtronic infusion set models for up to 8 days. Non-functional Medtronic pumps were used to simulate clinical use conditions for studying the adhesive patches on the infusion sets. The adhesive patches were evaluated for wear time (main objective), edge lift, and several other user feedback factors including: appearance; wear comfort; irritation score; adhesive residue upon removal; and removal discomfort. The Kaplan–Meier method was used to compare the retention (wear time) between adhesive materials.
Result:
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
Profil
Neuss, NRW, Germany
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
Profil
Neuss, NRW, Germany
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.