Researchers created a new artificial pancreas system called UniBE that uses advanced computer math to automatically deliver insulin for type 1 diabetes. In computer simulations, the system kept blood sugar in the healthy range 97.3% of the time under perfect conditions and 81.7% to 95.0% of the time when given realistic errors about meals and insulin doses, with almost no dangerous low blood sugar episodes. According to Gram Research analysis, this proof-of-concept study shows the system could outperform current insulin pumps, though it remains years away from human testing and real-world use.
Researchers at the University of Bern created a new artificial pancreas system called UniBE that uses advanced computer technology to automatically deliver insulin to people with type 1 diabetes. Unlike older systems that use simplified math models, UniBE uses a more sophisticated approach called successive linearization model predictive control to better predict how the body will respond to insulin. In computer simulations testing various real-world challenges like meal timing mistakes and carbohydrate counting errors, the system kept blood sugar in the healthy range 81.7% to 97.3% of the time while preventing dangerous low blood sugar episodes. This proof-of-concept study shows the system could work better than current automated insulin pumps, though it still needs real-world testing in patients before doctors can prescribe it.
Key Statistics
A 2026 computer simulation study of the UniBE hybrid automated insulin delivery system found that it maintained blood sugar in the target healthy range 97.3% of the time in baseline testing, compared to typical real-world performance of 70% to 85% for current commercial systems.
The UniBE system kept blood sugar in range between 81.7% and 95.0% across nine different test scenarios involving realistic errors in meal timing, carbohydrate counting, and insulin dosing, demonstrating resilience to clinically relevant disturbances.
In all testing scenarios, the UniBE system maintained dangerously low blood sugar (hypoglycemia) at near-zero rates, addressing a primary safety concern with automated insulin delivery systems.
The Quick Take
- What they studied: Can a new computer-controlled insulin pump system keep blood sugar levels more stable in people with type 1 diabetes by using better prediction math?
- Who participated: This was a computer simulation study, not a human trial. Researchers tested the UniBE system using a virtual patient model that mimics how real bodies respond to insulin.
- Key finding: In baseline testing, the system kept blood sugar in the healthy range 97.3% of the time. Even when the computer was given wrong information about meals or insulin doses, it still maintained blood sugar control between 81.7% and 95.0% of the time, with almost no dangerous low blood sugar episodes.
- What it means for you: This research suggests that future artificial pancreas systems could be smarter and more reliable than current pumps. However, this is early-stage research tested only in computers, not yet in real patients, so it will take several more years of testing before this system becomes available to use.
The Research Details
The UniBE system is a hybrid automated insulin delivery (hAID) system, which means it combines automatic insulin delivery with the ability for patients to manually give themselves insulin doses when needed. The researchers built a computer controller using a technique called successive linearization model predictive control. This fancy term means the system uses advanced math to predict what will happen to a patient’s blood sugar in the near future and adjusts insulin delivery accordingly.
The researchers tested their system using a computer simulation called the UVa/Padova metabolic simulator, which is an FDA-approved virtual patient model that realistically mimics how human bodies handle glucose and insulin. They ran nine different test scenarios to see how well the system worked when things went wrong in real life. These scenarios included mistakes in meal timing, errors in counting carbohydrates, and incorrect basal insulin doses (the steady background insulin that patients need all day).
The system was designed to work in two ways: it automatically delivers a steady background dose of insulin throughout the day, and it can also deliver extra insulin doses when needed for meals or to correct high blood sugar. The system adjusts these doses in real time based on continuous glucose monitor readings and can adapt to both automatic adjustments and doses that patients give themselves.
Current artificial pancreas systems use simplified math models to predict blood sugar changes, which can miss important details about how each person’s body is different. The UniBE system uses more advanced math that better captures these individual differences and can adapt more quickly to changes. This matters because better predictions mean better insulin dosing decisions, which leads to more stable blood sugar, fewer dangerous low blood sugar episodes, and less work for patients to manage their diabetes.
This study has important strengths and limitations to understand. The strength is that it was tested across nine different real-world error scenarios, showing the system is robust when things go wrong. The system was evaluated using an FDA-approved simulator, which adds credibility. However, this is purely a computer simulation study with no human participants. The researchers themselves note this is proof-of-concept only and does not constitute full pre-clinical validation. The study does not include real patient data, real insulin pumps, or actual glucose sensors. Before this system could ever be used by patients, it would need extensive testing in animals and then in human volunteers to confirm it is safe and actually works as well in real bodies as it does in computer simulations.
What the Results Show
In the baseline scenario where the computer had perfect information, the UniBE system achieved exceptional performance. It kept blood sugar in the target healthy range 97.3% of the time. The time spent with dangerously low blood sugar was essentially zero (0%), and time spent with high blood sugar was only 2.4%. These numbers are better than what most current artificial pancreas systems achieve in real-world use.
When the researchers introduced realistic errors and challenges, the system remained remarkably stable. Across all nine perturbation scenarios that included meal timing mistakes, carbohydrate counting errors, and basal insulin dose errors, the system maintained time in range between 81.7% and 95.0%. This means even when things went wrong, the system adapted and kept blood sugar mostly in the healthy range. Most importantly, the system maintained low rates of hypoglycemia (dangerous low blood sugar) across all scenarios, which is the primary safety concern with automated insulin delivery.
The system’s ability to adapt in real time to both automatic adjustments and patient-initiated insulin doses demonstrates that it could work well in real life where patients need some control over their therapy. The researchers found that the successive linearization approach allowed the system to handle the complex, nonlinear nature of glucose metabolism better than simpler linear models used in current systems.
The study demonstrated that the system’s performance remained stable even under multiple simultaneous errors, which is important because in real life, patients often make several mistakes at once (wrong meal time, wrong carb count, and incorrect basal dose). The system showed it could adjust its insulin delivery bounds in real time based on glucose dynamics, meaning it learned from what was happening and changed its strategy accordingly. This adaptability is a key feature that distinguishes it from less sophisticated systems.
Current commercial artificial pancreas systems like the Medtronic 780G and Tandem Control-IQ use linear or simplified models of glucose control. According to Gram Research analysis, the UniBE system’s use of successive linearization model predictive control represents an advancement in how these systems predict blood sugar changes. The reported time in range of 97.3% in baseline scenarios is higher than typical real-world performance of current systems, which usually achieve 70% to 85% time in range. However, direct comparison is difficult because this study used computer simulations while real-world systems are tested in actual patients with all the unpredictability that entails.
This study has several important limitations. First and most critically, it is entirely computer-based simulation with no human participants or real insulin pumps. The virtual patient model, while FDA-approved and realistic, cannot capture all the complexity of real human physiology. Second, the study does not include real continuous glucose monitor data or account for sensor errors, which are common in actual clinical use. Third, the sample size is zero because no real patients participated. Fourth, the study does not evaluate user experience, which is important because patients need to understand how to use the system. Fifth, the researchers explicitly state this is proof-of-concept only and does not constitute full pre-clinical validation. Finally, the study does not address how the system would perform over months or years of real-world use, or how it would handle unusual situations like illness, exercise, or hormonal changes that affect insulin needs.
The Bottom Line
Based on this computer simulation study, the UniBE system shows promise as a potential future treatment for type 1 diabetes. However, it is far too early to recommend it for actual patient use. The appropriate next steps are preclinical testing in animal models, followed by small human safety studies, and eventually larger clinical trials. People with type 1 diabetes should continue using their current approved insulin pumps and continuous glucose monitors, as these have been proven safe and effective in real patients. The confidence level for these recommendations is moderate, based on simulation data only.
This research is most relevant to people with type 1 diabetes who struggle to keep their blood sugar stable despite using current insulin pumps, and to their healthcare providers who are looking for better treatment options. It is also important for researchers and companies developing artificial pancreas technology. People with type 2 diabetes should not expect this system to be relevant to them, as it is specifically designed for type 1 diabetes where the pancreas produces little or no insulin. This research is not yet relevant to patients because the system is not available for clinical use.
If the UniBE system follows typical development timelines for medical devices, it would likely take 3 to 5 years of preclinical and clinical testing before it could potentially be submitted to the FDA for approval. If approved, it might become available to patients 5 to 10 years from now. This is a realistic timeline based on how long it takes to develop and test new insulin pump systems.
Frequently Asked Questions
What is an artificial pancreas system and how does it work?
An artificial pancreas is an automated insulin pump connected to a glucose sensor that works together like a real pancreas. The pump automatically delivers insulin based on blood sugar readings, adjusting doses throughout the day without requiring patients to manually calculate and inject insulin.
When will the UniBE system be available for patients to use?
The UniBE system is currently in early research stages tested only in computer simulations. It will require several years of animal testing and human clinical trials before the FDA could approve it. Realistic availability would be 5 to 10 years from now, if development continues successfully.
How is the UniBE system different from current insulin pumps like Medtronic or Tandem?
The UniBE system uses more advanced mathematical prediction models called successive linearization model predictive control, which better anticipates how blood sugar will change. This allows it to make smarter insulin dosing decisions than current systems that use simpler prediction methods.
Is this study proof that the UniBE system is safe and effective for patients?
No. This study only tested the system in computer simulations, not in real patients or with actual insulin pumps. The researchers explicitly state this is proof-of-concept only and does not constitute full pre-clinical validation. Real-world testing is still needed.
Could this system help people with type 2 diabetes?
No. The UniBE system is specifically designed for type 1 diabetes, where the pancreas produces little or no insulin. Type 2 diabetes involves different problems with insulin use and would require different treatment approaches.
Want to Apply This Research?
- Users could track their time in range (percentage of time blood sugar stays between 70 and 180 mg/dL) daily and weekly, along with the number of low blood sugar events and high blood sugar events. This mirrors the key metrics the UniBE system optimizes for.
- While this specific system is not yet available, current app users could prepare for future advanced systems by improving their carbohydrate counting accuracy and meal timing consistency. Users could practice logging meals at the time they eat them and learning the carbohydrate content of their favorite foods, skills that will be important when using more sophisticated automated systems.
- Long-term tracking should focus on glycemic variability (how much blood sugar bounces around) and time in range trends over weeks and months. Users should also track how often they need to manually correct high or low blood sugar, as this indicates how well the automated system is working. This data would be valuable to share with healthcare providers to assess whether a future system like UniBE is working well for that individual.
This research describes an experimental artificial pancreas system that has been tested only in computer simulations and is not yet available for patient use. The findings do not constitute clinical validation or FDA approval. People with type 1 diabetes should continue using their current approved insulin pumps and glucose monitoring systems as prescribed by their healthcare provider. Do not attempt to modify or replicate this system without proper medical supervision. Anyone interested in participating in future clinical trials of the UniBE system should consult with their endocrinologist or diabetes care team. This article is for educational purposes only and should not be used as a substitute for professional medical advice.
This research translation is published by Gram Research, the science division of Gram, an AI-powered nutrition tracking app.