Background and Aims

Children with T1D have varying insulin needs throughout a day due to factors including age, diurnal rhythms, exercise, food intake. This study investigates the variations in insulin needs for basal and auto-correction of children in different age groups, through MiniMed 780G data.

Methods

Pump and CGM data of 34 children using MiniMed 780G were obtained from Medtronic Carelink. Micro and auto boluses were analyzed on an hourly bases by two age groups as 5-9 and 9.1-18 years old. Glycemic metrics were analyzed based on the International CGM consensus.

Results

Mean age was 12.2 years, mean duration of diabetes was 6.1 years. A total of 4193 patient-days were analyzed. Mean TIR and GMI were 80.5%, 6.6%, respectively. Basal insulin ratio between 05am-07am is significantly higher than those between 10am-03am(p<0.01) whereas it was significantly lower between 07pm-09pm than those between 12am-10am(p<0.001)(Figure1).Auto-correction insulin ratio between 09pm-12am is significantly higher than those between 03am-05pm(p<0.001) and 07pm-09pm(p=0.008) whereas it was significantly lower between 07am-10am than those between 10am-03am(p<0.001). Basal insulin ratio is significantly higher in 5-9yo children than those among 9.1-18yo between 9pm-12am(p=0.026) and 12am-03am(p=0.003)(Figure2).

Figure1:Basal insulin/hour(%) for two age groups

Figure2:basal-insulin/hour(%) in time-intervals for two age groups

Conclusions

Minimed 780G data show that basal insulin needs are high in all age groups during the night and morning up to 10 am; also children under 9 years of age need more basal insulin around midnight, suggesting the reversed dawn phenomenon. Data obtained from AID systems can guide physicians to adjust insulin doses for MDI and initiation of conventional pumps.

Background and Aims

Glycaemic control in pediatric patients with Type 1 Diabetes (T1D) continues being a challenge. Insulin infusion systems are being developed that optimize and personalize insulin delivery.

The Medtronic MiniMed™780G is a new generation closed-loop hybrid system, that automatically adjusts insulin delivery and corrects glucose levels every five minutes to a modifiable target.

The aim of the study is to analyse glycaemic control data and glycaemic variability in pediatric patients with T1D after change from their usual treatment to the Medtronic™780G advanced close-loop system.

Methods

This is a prospective study in pediatric patients that begin treatment with the closed-loop system Minimed™780G, from different previous treatments. Data on glucose control and glycemic variability were studied at the beginning and 6 months after treatment.

Results

Twenty-eight patients (15 of them women) with a mean age of 13,5 years were studied. Four patients had previous treatment with MiniMed™640G system (sensor augmented pump with predictive low glucose suspend). The rest of the patients were in treatment with multiple daily injections, 19 of them associated continuous glucose monitoring with DEXCOM™G6 and the other 5 associated flash glucose monitoring with FREESTYLE2™.

A statistically significant reduction in HbA1c was observed, as well as an increase in time in range 70-180 mg/dl, a decrease in time in hyperglycemia and a reduction in time of hypoglycemia. An improvement in glycemic variability is also observed.

The results of the study are shown in table 1.

Conclusions

The new MiniMed™780G advanced closed-loop system improves metabolic control in pediatric patients with T1D, regardless of previous treatment.

Background and Aims

Single-hormone closed-loop treatment of type 1 diabetes is becoming more and more common. However, dual-hormone systems are still not available on the market. A dual-hormone artificial pancreas (AP) consists of 1) a continuous glucose monitor, 2) two pumps (one for insulin and one for glucagon) and 3) a control algorithm. The aim of this work is to develop a dual-hormone AP as well as to perform pre-trial in silico tests (of both the closed-loop system and the model identification).

Methods

The AP is based on nonlinear model predictive control (NMPC) and heuristics, and the model is identified using maximum likelihood estimation. In the pre-trial virtual clinical trial, we test the AP in closed-loop simulations of virtual persons. In reality, the AP does not know the true dynamics. Therefore, we generate the virtual individuals from a simulation model that is different from the control model in the AP.

Results

We present the results of closed-loop simulations on virtual persons using the dual-hormone AP as well as model identification. Furthermore, we demonstrate how it can be used to indicate if the system is ready for a real clinical trial. The system achieves time in range (TIR) (3.9 – 10 mmol/L) above the 70% TIR standard of care with a limited number of hypoglycemic events (< 3.9 mmol/L).

Conclusions

The pre-trial virtual clinical trial of the AP shows 1) time in range above the guideline-recommended target, 2) a limited number of hypoglycemic events, and 3) the safety heuristics function as expected and prevent undesired behavior.

Background and Aims

Inpatient use of fully-automated closed-loop insulin delivery has been shown in randomised clinical trials to be safe and improve glucose control compared with standard insulin therapy. This project investigates the feasibility of implementing the CamAPS HX closed-loop system for inpatients with type 2 diabetes in a tertiary hospital to inform widespread adoption.

Methods

An online training module hosted on the Cambridge Diabetes and Education Program (CDEP) platform was developed, and face-to-face or virtual workshops were used for staff training along with guidelines and policies for out-of-hours escalation.

Demographic and glycaemic data were collected using Electronic Patient Records. The project received local approval and was funded by Addenbrooke’s Charitable Trust.

Results

Ten healthcare professionals completed the online training module and 29 attended face-to-face or virtual training.

In the first 90 days of implementation, 12 inpatients (mean age 63±15 years, 92% male) with complex medical issues on ten wards started closed-loop insulin delivery with a total of 127 days of closed-loop usage. Patients had a mean 57.8±17.1% time in the target range (5.6 to 10.0mmol/L) and 37.9±17.7% time with glucose >10.0mmol/L during closed-loop use. There was 0.1% (0.0, 0.5) time spent in hypoglycaemia (<3.9mmol/L). Mean glucose was 9.7±1.2mmol/L. The median total daily insulin dose was 54 units/day (range 22 to 112 units/day). There were no episodes of severe hypoglycaemia or hyperglycaemic emergencies associated with use of the closed-loop system.

Conclusions

Implementation of fully-automated closed-loop systems in the hospital is feasible and can achieve safe and effective glucose control when delivered by the diabetes outreach team.

Background and Aims

Hybrid Closed Loop (HCL) systems improve time in range 70-180 mg/dL (TIR) but not all users meet the TIR target of ≥70%. This study developed a model to predict attainment of the consensus TIR target after 12-months of HCL use based on baseline and 1-month data among Tandem Control-IQ (CIQ) users.

Methods

Data from 162 youth (7.6±1.4 yrs., 45.7%F, 7.6%±1.4% HbA1c) who began using the CIQ HCL system were included. Lasso model selection was used to develop a predictive model for meeting the TIR goal after 12 months of use. The lasso is a single-step alternative to stepwise selection and includes covariates maximizing area under the curve (AUC) rather than using significance testing. Candidate factors included sex, age, diabetes duration, baseline HbA1c, race/ethnicity, insurance status, history of pump and continuous glucose monitor (CGM) use, and scores on psychosocial questionnaires, as well as percent CGM use, percent TIR, number of meal boluses/day, and percent HCL use at 1-month.

Results

Factors retained in the final model included 1-month TIR, meal boluses/day, and Hypoglycemia Fear Survey Helplessness/Worry About Low Blood Glucose score. The model had very good predictive ability with AUC of 0.84 (Figure). Internal 5-fold cross validation was also very good with an average AUC of 0.803±0.014.

Conclusions

Our prognostic model using clinically accessible baseline, early device-use and psychosocial data can strongly predict users who will meet therapeutic targets with HCL technology. This model may be useful in early identification of barriers so as to promote early intervention prior to behaviors becoming ingrained.

Background and Aims

Automated insulin delivery (AID) systems help people living with type 1 diabetes (T1D) obtain an optimal glycemic management (HbA1c ≤ 7%). However, not every AID user achieved this optimal target. We aim to investigate which characteristics are associated with an optimal glycemic management among adult AID users living with T1D.

Methods

Cross-sectional study using data from the BETTER registry, a registry recruiting participants living with T1D in Quebec, Canada. Inclusion criteria: T1D, aged ≥ 18 y/o, available HbA1c value and not pregnant. Participants were divided into HbA1c ≤ 7% group and HbA1c > 7% group. Student’s t test or chi-square test were used to compare the two groups. Multivariate logistic regression analysis was applied to analyze the associated factors.

Results

The 90 eligible participants (60.0% women) averaged (mean±SD) 43.5 ± 14.5 years old with 26.6 ± 12.5 years of T1D. Comparison between HbA1c ≤ 7% group (N=44) and HbA1c > 7% group (N=46) were shown in Table. Logistic regression analysis suggested that participants with bachelor degree or above (OR 4.18, 95%CI 1.45, 12.03) and with shorter duration of pump use (OR 1.12, 95%CI 1.02, 1.23) were more likely to attain an optimal glycemic management when using an AID, after adjusting for age, sex, body mass index and use frequency of AID (Figure).

Conclusions

Special attention should be given to adult AID users who have a lower educational level and a longer duration of pump use.

Background and Aims

To investigate the sustainability of glycemic improvements in youth using CIQ for one year

Methods

Youth with T1D starting CIQ enrolled in a observational study, stratified by baseline A1c: low (<7%), middle (7-8.9%), high (≥9%). Linear mixed models were used to compare changes from baseline to month-3 and month-3 to month-12.

Results

One-hundred eight-three youth (13±4 y, 52% M) were enrolled. Baseline TIR was 73±2%, 53±1% and 39±3% for the low, middle,and high groups, respectively, which increased to 79±2%, 67±1% and 53±2% at month-3, p<0.001 for all (Figure). TIR decreased at month-12 to 74±2% (p=0.003), 63±1% (p=0.006) and 49±3% (p=0.05) respectively. A1c in the high group decreased from 9.8±0.2% at baseline to 8.3±0.2% at month-3 (p<0.001), with no change at month-12 (p=0.74). In the middle group, A1c decreased from 7.7±0.1% at baseline to 7.2±0.1% at month-3 (p<0.001), then increased to 7.5±1.0% at month-12 (p=0.03). There was no change in A1c in the low group across time (6.3±0.1% to 6.4±0.1%, p=0.5 to 6.6±0.1%, p=0.18). Baseline meal boluses/day were 5.3±0.3, 4.5±0.3 and 2.6±0.6 for the low, middle and high groups respectively. Boluses/day decreased to 4.7±0.3 at month 3 (p=0.02) with no change at month 12 (p=0.45) in the low group. In the middle group, there was no change at month 3 (p=0.53), then a decrease to 3.7±0.3 at month 12 (p<0.001). There was no change in meal bolus frequency for the high group (p=0.90, 0.21).

Conclusions

Meal bolus frequency is an important factor to achieving glycemic targets with CIQ and sustaining glycemic improvements across time.

Background and Aims

Hybrid closed-loop (HCL) automated insulin delivery (AID) systems improve glycemic control in people with type 1 diabetes (T1D) but remain largely dependent on announcement and quantification of meals. Full closed loop (FCL) offers to remove this dependency but needs to be clinically validated.

Methods

In a randomized crossover supervised clinical trial of T1D adults, we assess the feasibility of three AID modalities: HCL, FCL, and FCL with anticipation of personalized meal patterns (FCL+). After a 4-week data collection, each modality was tested in random order during three identical 24h periods with standardized meals of different timing: dinner was 90-120min later than usual (per data collection); breakfast occurred as expected; lunch fixed at 1pm. We present an interim analysis (without comparative statistical analysis) of CGM-based outcomes (mean±sd or median[quartiles] as appropriate) overall, 2h-pre-meals, and 5h-post-meal.

Results

18 adult participants with T1D completed the protocol to date (N=36 expected at trial end). No serious adverse events were reported. Overall time in range (TIR) was excellent in all modalities (HCL: 86.3±8.2, FCL: 76.9±13.0%, FCL+: 78.1±12.1%), with low time below range (TBR, HCL: 0.7[0-2.8]%, FCL: 0[0-3.8]%, FCL+: 0.8[0-2.1]%). The nominal meal control (breakfast) showed HCL dominating FCL, with FCL+ in between: TIR=77.6±24.4%, 58.5±25.0%, and 66.3±23.6% respectively; increased insulin delivery pre-breakfast was observed for FCL+ (figure). Delaying meals showed no clear trend towards hypoglycemia, TBR, HCL:0[0-12.5]%, FCL:0[0-0]%, FCL+:0[0-8.3]%.

Conclusions

All modalities provided adequate glycemic control overall in this very controlled environment; meal anticipation appears to be safe and may mitigate some lost post-prandial control.

Background and Aims

Accurate modeling of glucagon effect is essential in dual-hormone artificial pancreas development, both for accurate in silico evaluations and the development of model-based control algorithms. Glucagon action models in literature differ significantly among authors and some of them do not provide physiological insight into glucagon behavior. This works aims to propose and validate a model based on glucagon receptors dynamics, which could justify some of the phenomena surrounding glucagon.

Methods

The proposed model was fitted using data from 8 persons with type 1 diabetes. Pharmacokinetic (PK) and pharmacodynamic (PD) related parameters were identified in such a way that the effect of glucagon on endogenous glucose production (EGP) was isolated. In order to provide a more insightful validation, three other models from literature were also fitted to data using the same procedure. A glucose-insulin-glucagon validated model from literature was used as common ground to test the glucagon model structures, only changing the glucagon effect description. This allowed to focus the comparison on the influence of glucagon on EGP.

Results

Results show that the average root mean square error of the fit with the proposed model was around 6.5 mg/dl. This value was 22.1% lower than the average error obtained with the other model structures.

Conclusions

The glucagon receptor dynamics enables the overall model to successfully fit clinical data. Therefore, this model will be useful in the development and design of dual-hormone artificial pancreas systems, as well as, providing a better understanding of glucagon physiology.

Background and Aims

In hybrid artificial pancreas systems (HAPs) insulin boluses are usually calculated based on patient estimations of the amount of carbohydrates to be ingested. The aim of this study is to calculate the bolus insulin without knowing the patient’s carbohydrate intake, thus alleviating the patient’s management burden.

Methods

A Q-Learning agent (QLA) was trained to optimize bolus insulin doses for in-silico type 1 diabetic patients. The area under the curve of glucose profile, maximum and minimum glucose values were defined as states. The glucose value before meal was utilized to define the range of bolus values in the action space to restrict the exploration of the QLA in a safe zone.

Results

The algorithm was tested for a cohort of 68 virtual patients and the results were compared to the standard bolus calculator (SBC) in open loop therapy. The results are given as median (interquartile range). A mean glucose value of 153.57 (145.54 - 166.88) vs 154.48 (145.52 - 164.21); p=0.0027, time below range of 0.049 (0.04 - 1.15) vs 1.17 (0.41 - 2.34); p=0.000000642, time in target range of 72.37 (59.99 - 81.94) vs 69.64 (61.56 - 77.40); p=0.0096 and time above range of 1.38 (0.27 - 4.68) vs 1.59 (0.7 - 4.44); p=0.645 were achieved for SBC and QLA respectively.

Conclusions

The reinforcement learning methodology using Q-Learning to compute insulin boluses without information on the amount of carbohydrates in meals showed similar performance as compared to the SBC.

Background and Aims

Little is known about the effect of long-term continuous/flash glucose monitoring (C/FGM) on hypoglycemia rates in the real world.

Methods

Online baseline data were obtained from a real-world panel of Americans (≥18 years old) with T1DM or T2DM taking insulin and/or secretagogues. Multivariable negative binomial regression was conducted to isolate the total effect of ≥1 year C/FGM use on self-reported, past-month non-severe and past-year severe hypoglycemia (NSH, SH). Confounding variables were identified from a directed acyclic graph.

Results

A complete case analysis was performed on 1,412 baseline responders (T1DM: 18.27%, age: 49.48 (SD: 14.16) years, male: 48.30%). One in ten (T1DM: 27.13%; T2DM: 7.28%) reported using a C/FGM device for ≥1-year preceding baseline. Overall, the crude rate of NSH and SH was 4.41 (95% CI: 4.30-4.52) events per person-month (EPPM) and 2.68 (95% CI: 2.60-2.77) events per person-year (EPPY), respectively. Those who used a C/FGM device for ≥1 year experienced 9.05 (95% CI: 8.58-9.54) non-severe EPPM and 5.63 (95% CI: 5.26-6.02) severe EPPY. Controlling for confounding, ≥1-year CGM users, versus non-C/FGM users, reported twice the number of past-year SH (2.05 [95% CI: 1.38-3.05, P<0.001]) and past-month NSH (2.02 [95% CI: 1.64-2.48, P<0.001]). Similar effects were observed among <1-year C/FGM users versus non-C/FGM users.

Conclusions

C/FGM can be a valuable tool to help detect and manage hypoglycemia; but, in and of itself, it is not a preventative treatment. Ongoing, vigilant clinical care of C/FGM users is still required to achieve hypoglycemia reduction in the real world.