Machine-Learning Insulin Delivery for Type 1 Diabetes

(AIDANET+BPS_RL Trial)

Overseen By Emma Emory, RN

Age: 18+

Sex: Any

Travel: May Be Covered

Time Reimbursement: Varies

Trial Phase: Academic

Recruiting

Sponsor: Boris Kovatchev, PhD

Must be taking: Insulin

Must not be taking: SGLT-2 inhibitors, Steroids

Disqualifiers: Seizure disorder, Severe hypoglycemia, DKA, others

No Placebo Group

Trial Summary

What is the purpose of this trial?

A randomized crossover trial assessing glycemic control using Reinforcement Learning trained Bolus Priming System (BPS_RL) added to the the Automated Insulin Delivery as Adaptive NETwork (AIDANET algorithm) compared to the original AIDANET algorithm.

Will I have to stop taking my current medications?

The trial requires that you do not start any new non-insulin glucose-lowering medications during the study. If you are currently using certain medications like SGLT-2 inhibitors or steroids, you may need to stop them before joining the trial.

What data supports the effectiveness of the treatment Bolus Priming System (BPS_RL) for type 1 diabetes?

Research on closed-loop insulin delivery systems, which automate insulin delivery, suggests they can improve blood sugar control in people with type 1 diabetes. These systems, similar to the Bolus Priming System, aim to mimic the body's natural insulin response, potentially reducing the need for manual insulin adjustments. ^{1 2 3 4 5}

Is the machine-learning insulin delivery system safe for humans?

The PEPPER system, which is similar to the machine-learning insulin delivery system, was found to be safe in a study with people who have type 1 diabetes. It includes safety features like alerts for low blood sugar and personalized insulin recommendations. ^{1 6 7 8 9}

How does the machine-learning insulin delivery treatment for type 1 diabetes differ from other treatments?

This treatment uses a machine-learning algorithm to automate insulin delivery, mimicking the natural insulin release of a healthy pancreas more closely than traditional methods. Unlike standard insulin pumps or injections, it can automatically adjust insulin doses based on real-time glucose measurements, potentially reducing the need for manual input and improving blood sugar control. ^{1 3 10 11 12}

Research Team

Sue Brown, MD

Principal Investigator

University of Virginia

Eligibility Criteria

This trial is for individuals with Type 1 Diabetes who are interested in improving their glycemic control. Specific eligibility criteria details were not provided, so it's important to contact the study organizers for more information on who can participate.

Inclusion Criteria

Agrees to use a form of contraception to prevent pregnancy while in the study

Willingness to use the study AIDANET system during the study period

I am 18 years old or older.

Exclusion Criteria

I have had a severe low blood sugar episode with seizure or fainting in the past year.

I have advanced kidney disease or am on dialysis.

My thyroid condition is not well-managed.

Trial Timeline

Screening

Participants are screened for eligibility to participate in the trial

2-4 weeks

Baseline Establishment

Participants use the AIDANET system at home for 7 days/6 nights to establish a baseline and initialize the control algorithm

1 week

Home use

Hotel Session

Participants are studied at a hotel session for 3 days/2 nights to assess glycemic control using the AIDANET or AIDANET+ BPS_RL systems

3 days

Hotel stay

Home Use Transition

Participants transition to home use of AIDANET+ BPS_RL for 7 days/6 nights

1 week

Home use

Follow-up

Participants are monitored for safety and effectiveness after treatment

4 weeks

Treatment Details

Interventions

• Bolus Priming System (BPS_RL) (Machine Learning)

Trial Overview The trial is testing a new system called AIDANET+ BPS_RL, which uses machine learning to help manage insulin delivery better than the current AIDANET algorithm alone. Participants will experience both methods in different periods to compare effectiveness.

Participant Groups

Group I: AIDANET→AIDANET+ BPS_RL Active Control 1 Intervention

Group A: AIDANET followed by AIDANET+ BPS_RL during the hotel session

Group II: AIDANET+ BPS_RL→AIDANET Active Control 1 Intervention

Group B: AIDANET+BPS_RL followed by AIDANET during the hotel session

Who Is Running the Clinical Trial?

Boris Kovatchev, PhD

Lead Sponsor

Trials

1

Recruited

120+

Sue Brown

Lead Sponsor

Trials

3

Recruited

100+

National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)

Collaborator

Trials

2,513

Recruited

4,366,000+

DexCom, Inc.

Industry Sponsor

Trials

151

Recruited

35,700+

Kevin Sayer

DexCom, Inc.

Chief Executive Officer since 2015

Bachelor’s and Master’s degrees in Accounting and Information Systems from Brigham Young University

Dr. Shelly Lane

DexCom, Inc.

Chief Medical Officer since 2023

MD from University of California, San Diego

Findings from Research

Hybrid closed-loop insulin delivery systems, known as artificial pancreas (AP), significantly improve glucose control in children and adolescents with type 1 diabetes, increasing the time spent in the target glucose range without raising the risk of hypoglycemia.

Despite the effectiveness of these systems for overnight glycemic control, daytime management remains challenging, highlighting the need for careful meal planning and bolusing, as fully automated systems are still under investigation.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Artificial Pancreas Technology Offers Hope for Childhood Diabetes. Schoelwer, MJ., DeBoer, MD.[2022]

A novel control system combining a bolus priming system (BPS) and a multistage model predictive control (MS-MPC) significantly improved post-meal blood glucose control in simulations, achieving the highest time in range at 60.73%.

All tested configurations maintained low hypoglycemia risk (time below 70 mg/dL <0.5%), indicating that the new system enhances glucose management without increasing the risk of low blood sugar.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Using an Online Disturbance Rejection and Anticipation System to Reduce Hyperglycemia in a Fully Closed-Loop Artificial Pancreas System. Corbett, JP., Garcia-Tirado, J., Colmegna, P., et al.[2022]

A decision tree model was developed using data from 196 adults with type 1 diabetes to predict the timing of insulin correction boluses, showing better classification performance than traditional logistic regression methods.

By embedding this model into a popular simulation tool, researchers can now conduct in-silico clinical trials that more accurately reflect real patient behaviors and their impact on blood glucose control after meals.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Generation of post-meal insulin correction boluses in type 1 diabetes simulation models for in-silico clinical trials: More realistic scenarios obtained using a decision tree approach. Camerlingo, N., Vettoretti, M., Del Favero, S., et al.[2022]

Stimuli-responsive, closed-loop insulin delivery systems can potentially improve glycemic control in type 1 diabetes by mimicking natural insulin release in response to changing glucose levels, reducing the burden of self-care for patients.

Although there are still challenges to overcome before these technologies can be used in clinical practice, advancements in reservoir-based insulin delivery devices show promise for significantly enhancing diabetes management.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Stimuli-Responsive Insulin Delivery Devices. Fuchs, S., Shariati, K., Ma, M.[2021]

The proposed automatic bolus and adaptive basal (ABAB) therapy for type 1 diabetes significantly outperformed traditional open-loop insulin therapy in a study involving 100 adult subjects, demonstrating better glucose control and reduced variability.

ABAB therapy showed a low hypoglycemia rate of only 3.3% even with increased insulin sensitivity, indicating its safety and robustness against insulin dosing mismatches, making it a promising candidate for future artificial pancreas systems.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Automatic bolus and adaptive basal algorithm for the artificial pancreatic β-cell. Wang, Y., Dassau, E., Zisser, H., et al.[2011]

The use of an automated bolus advisor (ABA) in adults with type 1 diabetes led to a reduction in glycated hemoglobin (HbA1c) levels, indicating improved glycemic control, although the difference compared to traditional methods was not statistically significant.

Patients using the ABA reported a lower rate of postprandial hypoglycemia, increased daily glucose measurements, and higher satisfaction and quality of life, suggesting that the ABA is a well-accepted and effective tool for managing diabetes.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Expert Study: Utility of an Automated Bolus Advisor System in Patients with Type 1 Diabetes Treated with Multiple Daily Injections of Insulin-A Crossover Study. Gonzalez, C., Picón, MJ., Tomé, M., et al.[2017]

The PEPPER system, which provides personalized bolus advice for type 1 diabetes management, was found to be safe but did not significantly improve the percentage of time in range (TIR) compared to a standard bolus calculator, with TIR percentages of 62.5% for PEPPER and 58.4% for control.

Participants using the PEPPER system reported a higher perception of hypoglycemia, even though there was no objective difference in actual time spent in hypoglycemia, indicating a need for further studies to assess the system's overall effectiveness and user experience.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Safety and Feasibility of the PEPPER Adaptive Bolus Advisor and Safety System: A Randomized Control Study. Avari, P., Leal, Y., Herrero, P., et al.[2022]

A modular safety system integrated into an intelligent insulin dose recommender significantly improved glycemic control in a pilot study with eight adults over six weeks, reducing time spent in hypoglycemia from 0.82% to 0.33%.

The system also increased the percentage of time participants maintained their blood glucose within the target range (3.9-10.0 mmol/l) from 52.8% to 61.3%, demonstrating its efficacy in enhancing safety during insulin dosing.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

A Modular Safety System for an Insulin Dose Recommender: A Feasibility Study. Liu, C., Avari, P., Leal, Y., et al.[2021]

The study developed a model predictive control (MPC) algorithm for an automated insulin delivery system that significantly improves blood glucose control in Type 1 diabetes by incorporating an insulin on board (IOB) safety constraint.

Simulations showed that using the IOB constraint reduced the risk of hypoglycemic events from 50% to 10%, highlighting its importance in ensuring both safety and efficacy in insulin therapy.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Safety constraints in an artificial pancreatic beta cell: an implementation of model predictive control with insulin on board. Ellingsen, C., Dassau, E., Zisser, H., et al.[2021]

The study developed algorithms that significantly improved blood glucose control for people with type 1 diabetes using multiple-daily-injections, increasing the time spent in the target glucose range from 66.9% to 93.6% over 20 weeks.

These algorithms adapt insulin delivery based on self-monitoring of blood glucose, suggesting that automated decision support systems can enhance glycemic regulation and reduce the management burden for patients.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Using Iterative Learning for Insulin Dosage Optimization in Multiple-Daily-Injections Therapy for People With Type 1 Diabetes. Cescon, M., Deshpande, S., Nimri, R., et al.[2021]

The development of an artificial pancreas aims to closely mimic the natural insulin secretion of healthy individuals, focusing on replicating both first-phase and second-phase insulin responses to maintain normal glucose levels.

Recent advancements in algorithms and technology have improved the effectiveness of closed-loop systems that use glucose sensors and insulin pumps, showing promising results in managing insulin-dependent diabetes.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Closed-loop insulin delivery-the path to physiological glucose control. Steil, GM., Panteleon, AE., Rebrin, K.[2019]

References

1. United States pubmed.ncbi.nlm.nih.gov

Six-Month Randomized, Multicenter Trial of Closed-Loop Control in Type 1 Diabetes. [2022]

2. United States pubmed.ncbi.nlm.nih.gov

Artificial Pancreas Technology Offers Hope for Childhood Diabetes. [2022]

3. United States pubmed.ncbi.nlm.nih.gov

Using an Online Disturbance Rejection and Anticipation System to Reduce Hyperglycemia in a Fully Closed-Loop Artificial Pancreas System. [2022]

4. Ireland pubmed.ncbi.nlm.nih.gov

Generation of post-meal insulin correction boluses in type 1 diabetes simulation models for in-silico clinical trials: More realistic scenarios obtained using a decision tree approach. [2022]

5. United States pubmed.ncbi.nlm.nih.gov

Stimuli-Responsive Insulin Delivery Devices. [2021]

6. United States pubmed.ncbi.nlm.nih.gov

Automatic bolus and adaptive basal algorithm for the artificial pancreatic β-cell. [2011]

7. United States pubmed.ncbi.nlm.nih.gov

Expert Study: Utility of an Automated Bolus Advisor System in Patients with Type 1 Diabetes Treated with Multiple Daily Injections of Insulin-A Crossover Study. [2017]

8. United States pubmed.ncbi.nlm.nih.gov

Safety and Feasibility of the PEPPER Adaptive Bolus Advisor and Safety System: A Randomized Control Study. [2022]

9. United States pubmed.ncbi.nlm.nih.gov

A Modular Safety System for an Insulin Dose Recommender: A Feasibility Study. [2021]

10. United States pubmed.ncbi.nlm.nih.gov

Safety constraints in an artificial pancreatic beta cell: an implementation of model predictive control with insulin on board. [2021]

11. United States pubmed.ncbi.nlm.nih.gov

Using Iterative Learning for Insulin Dosage Optimization in Multiple-Daily-Injections Therapy for People With Type 1 Diabetes. [2021]

12. Netherlands pubmed.ncbi.nlm.nih.gov

Closed-loop insulin delivery-the path to physiological glucose control. [2019]