Impact of erroneous meal insulin bolus with dual-hormone artificial pancreas using a simplified bolus strategy – A randomized controlled trial


Study design

The first part of this study is a single-blind, randomized, two-way, cross-over study to examine the impact of overestimation of a meal insulin bolus in the context of dual-hormone closed-loop operation (insulin and glucagon) combined with a simplified semi-quantitative meal-size estimation in adults with type 1 diabetes. Two separate meals (75 g and 45 g of carbohydrate) were tested in clinical settings during two subsequent clinical trials registered as NCT02626936 (December 10, 2015) and NCT02798250 (June 14, 2016). The second part of this study involved in silico clinical trials to explore the effect of an underestimated, adequately estimated, and overestimated boluses with three different meal sizes (45 g, 75 g, and 105 g of carbohydrate). These trials mimicked the clinical experiments in duration and protocol. The in silico trials were conducted using the same dual-hormone dosing predictive algorithm that was used for the human clinical trials30. The algorithm is based on a fuzzy-supervised model-based predictive technique combined with extended Kalman filtering and a set of heuristic rules. The algorithm was initialized using body weight, daily insulin requirements and insulin-to-carbohydrate ratios. Glucagon delivery is based on heuristic logical rules employing estimates of plasma glucose concentrations and their trends as provided by the Kalman filter and is accompanied with suspension of insulin delivery and reduced aggressiveness of insulin dosing. The in silico trials were conducted using the Development Platform for Artificial Pancreas Algorithms system31.

Study population

Participants were recruited and tested from January 2016 to March 2017 at the Institut de Recherches Cliniques de Montréal (IRCM), Montréal, Canada. We included adults (≥18 years of age) with type 1 diabetes for >one year, using insulin pump therapy for at least three months and using carbohydrate counting for meal boluses. Participants with poor glucose control (glycated hemoglobin >10%), with clinically significant micro (e.g. gastroparesis) or macrovascular complications or using medication likely to affect the results interpretation (e.g., agents affecting gastric emptying) were excluded. Other exclusion criteria were applied as detailed in the clinical trial registry. Ten participants were recruited for each study (75 g and 45 g of carbohydrate meals). Patients were offered to participate to one or both studies and four patients participated to both experiments. The studies obtained ethical approval from the IRCM ethics committee and all participants provided informed written consent. All experiments were performed in accordance with relevant guidelines and regulations. The simulated trials were conducted with 15 virtual patients as described below.

Study procedures

All participants first completed a screening visit to assess eligibility. This visit included a medical visit with an endocrinologist to confirm eligibility and included anthropometric measurements (height, weight, waist circumference), a blood draw (glycated hemoglobin), and examination of insulin therapy records (average prandial and basal doses, basal rates, insulin-to-carbohydrate ratios). Participants were instructed how to install and use the continuous glucose monitoring systems (Dexcom G4 Platinum, Dexcom) or they could also set an appointment with study personnel for installation. They were instructed to calibrate sensors at least twice, and maximum four times daily using capillary blood glucose values.

Interventions were undertaken one to five days after sensor insertion. Subjects were admitted at IRCM at 7:00 am and were asked to fast from the previous night (midnight). If a correction bolus or a hypoglycemia correction (food or carbohydrate intake) was performed by patients within the four hours preceding the test, the test was reported or delayed depending on the magnitude of treatment or correction. Upon arrival, patients’ pumps were substituted with the study pump and a second pump containing glucagon was installed. During the interventions, variable subcutaneous insulin (patient’s usual insulin: Lispro, Aspart or Glulisine) and glucagon (Eli Lilly) infusion rates were used to regulate postprandial glucose levels using two infusion pumps (Accu-Check Combo, Roche). Glucagon was reconstituted according to the manufacturer’s instructions. Capillary blood glucose was measured to calibrate the sensor. At 7:00 am, closed-loop delivery was initiated and a standardized breakfast (45 g or 75 g of carbohydrate) was served at 9:00 am. Subjects were offered a choice between two menus (Appendix A), but consumed the same meal on both interventions. During the intervention, patients were allowed to do sedentary activities only (reading, watching television, playing video games, etc.). They were also allowed coffee or tea, but their consumption had to be replicated during the second intervention. The intervention ended at 1:00 pm. Interventions were scheduled 0 to 14 days apart.

The glucose levels as measured by the real-time sensor were entered manually into the computer every 10 minutes. The pumps’ infusion rates were then changed manually, using a remote control device, based on the computer-generated recommendation from the predictive algorithm30. The simplified bolus strategy is based on the estimated meal size (regular, large or very large) and the closed-loop delivery system gives the remaining insulin needed based on the sensor readings. Prandial boluses are calculated as individualized insulin-to-carbohydrate ratios multiplied by a fixed carbohydrate factor depending on the category. A regular meal is defined as 30 to 60 g of carbohydrate, with a bolus for 35 g. A large meal is defined as 60 to 90 g of carbohydrate, with a bolus for 65 g. A very large meal is defined as more than 90 g, with a bolus for 95 g. Patients thus have to select a category for their meal rather than count the exact amount of carbohydrates. For the purpose of this study, the meal category was entered by study personnel based on the randomized sequence (adequate estimation or overestimation): for the 75 g of carbohydrate meal, a bolus for 65 g (adequate estimation: large meal) or 95 g (overestimation: very large meal) was provided and for the 45 g of carbohydrate meal, a bolus for 35 g (adequate estimation: regular meal) or 65 g (overestimation: large meal) was provided. Participants were blinded to the intervention but not study personnel. To test the effect of overestimation on very large meals, we simulated in silico an additional trial with a 105 g of carbohydrate meal with a bolus for 65 g (underestimation: large meal), a bolus for 95 g (adequate estimation: very large meal), and a bolus for 125 g (overestimation: overbolus, undefined meal category). Insulin boluses were administered 10 minutes before the beginning of the meal. A balanced randomization, created by a third-party unrelated to the project was used to determine the order of the interventions. Randomization allocation was preserved in sealed envelopes opened by the study personnel following admission.

Simulated Trials

To augment the clinical trials in exploring the impact of meal bolus overestimation using the simplified semi-quantitative strategy within a closed-loop system, we used the cloud-based Development Platform for Artificial Pancreas Algorithms31 system to conduct the simulated clinical trials. The system was connected with the same predictive algorithm30 used in the trials and followed the same clinical protocols (4-hour post-meal period, meals at 9 am, and dosing amounts calculated every 10 minutes). The simulated trials were performed on 15 virtual subjects whose glucoregulatory responses were modeled from the responses of real type 1 diabetes patients24. The patient model incorporates inter- and intra-individual variability in patient’s glucose, insulin, and glucagon dynamics.


The primary endpoint was the percentage of time with sensor glucose below 4.0 mmol/L for the 4-hour period following the meal compared between dual-hormone closed-loop delivery with adequate estimation (proper categorization) vs. dual-hormone closed-loop delivery with overestimation (erroneous categorization). Secondary endpoints were also calculated for the 4-hour period following the meal and included; mean sensor glucose, percentage of time for which sensor glucose was in the target range (4.00–10.00 mmol/L), percentage of time spent above the target range, peak and time to peak sensor glucose, insulin bolus, total amount of insulin delivered, and the total amount of glucagon delivered.

Statistical analyses

The effect of the interventions on the continuous outcomes was estimated using a multivariate linear mixed effect model (LMEM) with the intervention (2 levels), treatment sequence, period, and starting glucose level entered as fixed effects, and subject nested within sequence as random effect. If highly skewed (P value for Shapiro–Wilk test < 0.05), outcomes were transformed using an appropriate Box-Cox transformation prior to modeling the linear mixed model. This is an exploratory study and power calculations were not performed. Data are presented as median [Interquartile Range].

All data generated or analysed during this study are included in this published article (and its Supplementary Information files).

Clinical trial registry numbers

NCT02626936 (December 10, 2015), NCT02798250 (June 14, 2016).

Source link

Products You May Like

Articles You May Like

Tattoo removal cream: Does it work?
Is Rituximab Effective for Multiple Sclerosis? – Medical News Bulletin
New Advances in Lung Cancer Treatment – Medical News Bulletin
Glioblastoma: Temozolomide, Bevacizumab May Improve Outcomes Among Elderly Patients
Phase 2 Study of Chemoradiotherapy Alone or With Surgery for Esophageal and Gastric Cancer

Leave a Reply

Your email address will not be published. Required fields are marked *