Automated Insulin Delivery: From Pancreas to Algorithm

The Journey So Far: In Parts 1 and 2, we explored the physiological foundations of glucose control and the mathematical modeling that enabled virtual testing of automated insulin delivery systems. The UVA/Padova simulator's FDA acceptance in 2008 opened the floodgates for innovation, leading to a diverse ecosystem of AID systems each employing distinct control strategies. In this concluding installment, we examine the major commercial systems that have reached patients, the remarkable DIY movement that democratized access, and the emerging role of incretins that may fundamentally reshape type 1 diabetes management.

The Commercial AID Landscape: From Simulation to Clinical Reality

TypeZero Technologies: The Direct Heir to UVA/Padova

inControl Platform

Founded in 2010 by researchers directly involved in creating the UVA/Padova simulator—including Boris Kovatchev and colleagues—TypeZero Technologies represented the most direct translation of academic research into commercial products. The company's inControl platform employed a zone-based model predictive control algorithm that divided glucose management into distinct operational zones.¹

Novel Features: The zone-MPC approach defined safety zones (preventing hypoglycemia), correction zones (addressing hyperglycemia), and target zones (maintaining euglycemia). This architecture allowed the algorithm to behave differently based on glucose trajectory, being conservative when approaching hypoglycemia but more aggressive when correcting hyperglycemia. The system incorporated adaptive learning that personalized insulin sensitivity parameters over time, meal detection algorithms that could identify unreported carbohydrate intake, and asymmetric cost functions that heavily penalized hypoglycemia relative to hyperglycemia.²

Key Strengths

Direct lineage from validated UVA/Padova simulator
Proven clinical trial data showing excellent safety profile
Sophisticated adaptation to individual physiology
Strong hypoglycemia prevention

Limitations

Required meal announcements (hybrid closed-loop)
Conservative approach sometimes limited time-in-range during postprandial periods
Computational complexity required significant processing power

Clinical Impact: TypeZero's technology was acquired by Dexcom in 2018, and its algorithms now power several commercial AID systems. The inControl platform demonstrated time-in-range of 70-75% in pivotal trials with severe hypoglycemia rates below 1 event per patient-year, establishing benchmarks for the field.³

UC Santa Barbara / Sansum Diabetes Center: Model Predictive Control Excellence

The Doyle-Zisser MPC Algorithm

Frank Doyle III and Howard Zisser led development of one of the most mathematically sophisticated MPC approaches for automated insulin delivery. Their work at UCSB's Artificial Pancreas Center, in collaboration with Sansum Diabetes Research Institute, pushed the boundaries of what predictive control could achieve.⁴

Novel Features: The UCSB approach employed a velocity-form MPC that predicted glucose trajectories 60-90 minutes into the future using a linear time-varying model identified in real-time from CGM data. Unlike model-based approaches requiring extensive patient-specific parameters, this algorithm learned glucose dynamics adaptively from recent history. The system featured zone-based target adjustment that relaxed glucose targets during exercise or illness, constraint handling that explicitly limited insulin delivery rates to prevent hypoglycemia, and disturbance rejection capabilities that accounted for unannounced meals and physical activity.⁵

Key Strengths

Minimal patient-specific tuning required
Rapid adaptation to changing conditions
Excellent handling of exercise and activity
Mathematically rigorous approach with stability guarantees

Limitations

Performance degraded with very irregular meal patterns
Required consistent CGM data quality
Linear model assumptions sometimes limiting for extreme conditions

Clinical Translation: While not directly commercialized as a standalone product, the UCSB MPC approach influenced multiple commercial systems and established MPC as the dominant paradigm for AID control algorithms. Clinical studies demonstrated 72-78% time-in-range with particularly strong overnight control.⁶

CamAPS FX: Cambridge's Adaptive Algorithm

Roman Hovorka's Adaptive Control System

The Cambridge group led by Roman Hovorka developed one of the most clinically successful AID systems, combining sophisticated modeling with pragmatic clinical design. CamAPS FX (Cambridge Adaptive Pancreas System) received CE marking in Europe in 2020 and FDA approval in the United States in 2022.⁷

Novel Features: CamAPS FX employs a nonlinear MPC algorithm based on Hovorka's extensively validated glucose-insulin model first published in 2004. The system's distinctive feature is its adaptive glucose target that automatically adjusts based on recent glucose patterns and user behavior. The algorithm learns from meal announcements and responses, gradually improving carbohydrate counting accuracy suggestions. Real-time parameter estimation continuously updates insulin sensitivity, with circadian patterns recognized automatically. An optional "boost" mode provides more aggressive correction for persistent hyperglycemia while maintaining safety constraints.⁸

Key Strengths

Highly adaptive to individual patterns
User-friendly interface with minimal required inputs
Extensive pediatric validation and approval
Proven real-world effectiveness across diverse populations

Limitations

Still requires meal announcements for optimal performance
Adaptation period of 1-2 weeks for new users
Performance varies with carbohydrate estimation accuracy

Clinical Success: The pivotal trial of CamAPS FX showed time-in-range improvement from 61% to 72% compared to sensor-augmented pump therapy, with mean HbA1c reduction of 0.4%. Importantly, the system demonstrated consistent performance across ages 1-70 years, making it one of the most versatile AID systems available.⁹

DreaMed / Medtronic: Fuzzy Logic and Clinical Integration

MD-Logic and Guardian Connect Integration

Moshe Phillip and colleagues at the Schneider Children's Medical Center in Israel developed the MD-Logic artificial pancreas system, which Medtronic later licensed and integrated into its diabetes management ecosystem. This system took a distinctly different approach from pure MPC algorithms.¹⁰

Novel Features: MD-Logic employs fuzzy logic control rather than traditional MPC, using rule-based decision-making that mimics clinical reasoning patterns. The system defines fuzzy sets for glucose levels (very low, low, target, high, very high) and rates of change, then applies clinical decision rules developed from expert endocrinologist input. The algorithm incorporates time-of-day specific rules recognizing different insulin sensitivity patterns at different times, probabilistic meal detection without requiring announcements, and safety systems that aggressively prevent predicted hypoglycemia.¹¹

Medtronic's implementation in the MiniMed 780G system added SmartGuard technology with auto-correction boluses every five minutes when glucose exceeded targets, integration with Guardian 4 sensor eliminating calibrations, and Bluetooth connectivity enabling smartphone-based monitoring and control.¹²

Key Strengths

Intuitive clinical logic easily understood by providers
Excellent overnight glucose control
Automatic correction boluses reduce burden
Robust performance with missed meal announcements

Limitations

Frequent auto-corrections can lead to overinsulinization in some users
Less predictive than MPC approaches
Performance depends heavily on accurate insulin-to-carb ratios
Proprietary ecosystem limits sensor and pump choices

Market Position: The MiniMed 780G represents the latest generation of Medtronic's AID systems, demonstrating 75-80% time-in-range in real-world studies. Its fully integrated approach appeals to users preferring a single-vendor solution, though this integration limits component flexibility.¹³

Beyond the Major Players: Ecosystem Diversity

Other Commercial Systems

The AID landscape includes additional commercial offerings, each with unique characteristics. Tandem Diabetes Care's Control-IQ technology, developed in partnership with TypeZero/Dexcom, uses a model-free adaptive algorithm with predictive low glucose suspend and automatic correction boluses. Insulet's Omnipod 5 employs an adaptive MPC algorithm in a tubeless patch pump format, offering unique form factor advantages. Beta Bionics' iLet bionic pancreas uses a fully autonomous approach requiring only body weight as input, with both insulin-only and bihormonal configurations under development.^14,15

The DIY Revolution: Democratizing Innovation

OpenAPS: The Movement That Changed Everything

Perhaps the most remarkable chapter in AID history came not from academic labs or corporate R&D departments, but from patients themselves. In 2013, when no commercial AID systems were available, Dana Lewis and Scott Leibrand developed OpenAPS (Open Artificial Pancreas System), creating the first publicly available DIY closed-loop system.¹⁶

The DIY Ethos: The OpenAPS movement embodied a fundamental shift in medical device development—patients unwilling to wait for regulatory approval created their own solutions using commercially available components. The community openly shared algorithms, troubleshooting guides, and safety protocols, prioritizing patient autonomy and rapid innovation over formal regulatory pathways.

OpenAPS uses a relatively simple algorithm compared to commercial MPC systems, but its simplicity proved advantageous for safety and transparency. The system calculates insulin delivery based on current glucose, glucose trend, insulin on board, carbohydrate absorption estimates, and safety constraints that prevent excessive insulin delivery. Users can inspect and modify all algorithm parameters, enabling unprecedented personalization.¹⁷

Loop and AndroidAPS: Expanding the Ecosystem

The DIY movement expanded with Loop (for iOS devices) and AndroidAPS, each bringing different technical approaches and user communities. Loop employs an algorithm developed by Nate Racklyeft using retrospective glucose data to predict future values, while AndroidAPS implemented OpenAPS algorithms on Android devices with additional customization options.¹⁸

These systems demonstrated that sophisticated AID could be implemented on consumer smartphones, that open-source development could maintain high safety standards through community peer review, and that patient-driven innovation could achieve outcomes comparable to commercial systems in motivated users.

Clinical Evidence and Regulatory Perspectives

Remarkably, DIY systems have been studied in formal clinical research. The CREATE trial published in 2022 demonstrated that adults using DIY AID systems achieved time-in-range of 71% compared to 59% with standard therapy, with similar safety profiles to commercial systems. However, DIY systems require technical sophistication, constant vigilance, and willingness to accept responsibility for unapproved device use.¹⁹

Regulatory agencies have responded with cautious acknowledgment. The FDA has stated it neither approves nor recommends DIY systems but recognizes patient autonomy in device choices. This pragmatic stance reflects the reality that thousands of patients successfully use DIY systems while commercial options continue expanding.

The Incretin Revolution: Beyond Insulin Alone

GLP-1 Receptor Agonists Enter the Type 1 Diabetes Arena

As AID systems achieved maturity, an unexpected therapeutic avenue emerged from the type 2 diabetes world. Glucagon-like peptide-1 (GLP-1) receptor agonists, originally developed for type 2 diabetes, began appearing in type 1 diabetes management through off-label use by adventurous clinicians and patients.²⁰

GLP-1 agonists offer several mechanisms potentially beneficial in type 1 diabetes beyond their insulin-independent effects. They slow gastric emptying, reducing postprandial glucose excursions and potentially simplifying insulin dosing timing. They suppress glucagon secretion, which is often dysregulated in type 1 diabetes. They promote satiety and weight loss, addressing weight gain often associated with intensive insulin therapy. They may preserve residual beta-cell function in early-stage disease, though this remains controversial.²¹

Dual Agonists: Tirzepatide's Promise

The latest frontier involves dual GLP-1/GIP (glucose-dependent insulinotropic polypeptide) receptor agonists, particularly tirzepatide. Approved for type 2 diabetes in 2022 with remarkable efficacy, tirzepatide quickly gained attention in the type 1 diabetes community for off-label use.²²

Emerging Clinical Data: Early case series and small studies of GLP-1 agonists in type 1 diabetes showed HbA1c reductions of 0.3-0.6% when added to insulin therapy, weight loss of 3-7 kg over 6 months, reduced total daily insulin requirements by 10-20%, and improved postprandial glucose control. However, these studies also revealed increased risk of gastrointestinal side effects and concerns about diabetic ketoacidosis despite normal or mildly elevated glucose levels.²³

Lilly's Groundbreaking Type 1 Diabetes Trials

Recognizing the potential, Eli Lilly initiated formal clinical trials of tirzepatide in type 1 diabetes populations. The SURPASS-T1D program represents the first large-scale investigation of a dual incretin agonist in type 1 diabetes, with trials evaluating safety, efficacy, and optimal dosing strategies in adults and adolescents with established type 1 diabetes.²⁴

These trials address critical questions including the optimal tirzepatide dose for type 1 diabetes, which may differ from type 2 diabetes dosing, strategies for insulin dose reduction when initiating incretin therapy, risk mitigation for diabetic ketoacidosis, and long-term effects on residual beta-cell function and complications.

Implications for AID Systems

The integration of incretins with AID systems represents a paradigm shift in type 1 diabetes management. AID algorithms will need adaptation to account for slower carbohydrate absorption with GLP-1 therapy, potentially enabling reduced meal boluses. Ketoacidosis prevention requires new monitoring strategies and alert systems, as traditional glucose-based warnings may be insufficient. The combination therapy might achieve time-in-range exceeding 80%, approaching near-normal glucose control.²⁵

Several research groups are developing hybrid algorithms that adjust insulin delivery based on incretin therapy presence. These systems recognize the altered glucose-insulin dynamics and modify prediction models accordingly, potentially offering the best glycemic control yet achieved.

Challenges and Limitations Across All Systems

Persistent Technical Challenges

Despite remarkable progress, all current AID systems face common limitations. Subcutaneous insulin delivery inherent delays cannot be eliminated with current technology. CGM accuracy issues persist, particularly during rapid glucose changes and in the hypoglycemic range. The meal announcement burden remains significant in most systems, even those claiming fully closed-loop operation. Exercise management requires user intervention and algorithm adjustments in almost all systems.²⁶

Psychosocial and Economic Barriers

Beyond technical limitations, substantial barriers limit AID access. The cost of AID systems ranges from eight thousand to fifteen thousand dollars annually in the United States, often with significant insurance barriers. Technical complexity and alarm fatigue can overwhelm users, particularly those with limited digital literacy. Healthcare provider education lags behind technology development, with many clinicians uncomfortable managing AID therapy. Socioeconomic disparities create troubling access gaps, with AID use predominantly among higher-income populations.²⁷

Future Directions: The Next Decade of Innovation

Near-Term Advances

The next few years will likely bring ultra-rapid insulin analogs with onset of action under five minutes, potentially reducing the meal announcement burden. Improved CGM accuracy with mean absolute relative difference below five percent will enable more aggressive algorithms. Machine learning integration will allow algorithms that learn individual patterns without explicit programming. Smartphone-based control will become universal, eliminating dedicated controller devices.²⁸

Medium-Term Possibilities

Looking five to ten years ahead, more speculative advances include bihormonal systems incorporating glucagon delivery for superior hypoglycemia prevention and correction, implantable sensors and pumps eliminating external hardware, and closed-loop systems requiring no meal announcements through sophisticated meal detection and predictive algorithms. Incretin-insulin combination systems will likely emerge with integrated dosing algorithms optimized for dual therapy.²⁹

The Ultimate Goal: Cure and Prevention

Technology advances, while transformative, remain symptomatic treatment. The ultimate goals remain beta-cell replacement through stem cell-derived islets or encapsulated cell therapies, and immunotherapy to prevent or reverse autoimmune destruction. Until these curative approaches succeed, AID systems represent the best available therapy for achieving near-normal glucose control with acceptable burden.³⁰

Conclusion: A Transformed Landscape

The journey from Bergman's minimal model to today's sophisticated AID systems spans four decades of iterative innovation. What began as mathematical abstractions in academic papers has materialized into commercial products and DIY solutions improving lives daily for tens of thousands of people with type 1 diabetes.

Each system discussed—from TypeZero's zone-MPC to Cambridge's adaptive algorithm to DreaMed's fuzzy logic—contributed unique insights to the collective understanding of automated glucose control. The DIY community demonstrated that innovation need not follow traditional pathways, and that patient agency can drive progress alongside corporate development. The emerging incretin story suggests we may be on the cusp of another paradigm shift, moving beyond insulin-only therapy to multi-hormone approaches.

Yet significant work remains. Access disparities must be addressed through cost reduction and policy changes. Technical limitations require continued research on faster insulins, more accurate sensors, and smarter algorithms. The psychosocial burden of diabetes persists even with automation, requiring attention to mental health and quality of life.

For now, automated insulin delivery has transformed type 1 diabetes from a condition requiring constant manual intervention to one where technology handles most routine decisions, freeing patients to live with reduced burden and improved outcomes. As systems continue improving and new therapeutic modalities emerge, the vision of truly worry-free diabetes management draws ever closer to reality.