Closed-Loop Insulin Delivery Algorithm Landscape 2026
Closed-Loop Insulin Delivery Algorithm Landscape 2026
From PID controllers to compound rMPC/rPID architectures and LSTM-based prediction, closed-loop insulin delivery algorithms have become the primary competitive differentiator in the artificial pancreas market. This dataset spans 2009–2026, covering 12 formal patent records and more than 50 clinical literature records.
How Closed-Loop Insulin Delivery Algorithms Have Evolved
Closed-loop insulin delivery (CLID) integrates continuous glucose monitoring (CGM), a subcutaneous insulin pump, and a control algorithm to automate glycemic regulation. The core algorithmic challenge is translating real-time interstitial glucose measurements into precise insulin dosing commands while treating hypoglycemia avoidance as a hard safety constraint.
The dataset spans publication dates from 2009 to 2026 and contains at least 6 distinct formal patent filings alongside more than 50 clinical and research literature records. The field has evolved from single-loop PID controllers toward multi-layered predictive, adaptive, and learning-based architectures across this period.
Key sub-domains visible in the data include PID and risk-transformed PID variants, model predictive control (MPC) and zone-MPC, iterative and reinforcement learning controllers, hybrid compound algorithms, open-source community-derived systems, dual-hormone approaches, and exercise- and meal-aware contextual adaptations.
In this dataset, Medtrum Technologies Inc. is the single most prolific patent filer with 8 WO/US filings in 2021–2023, followed by Tandem Diabetes Care with 3 filings and Medtronic MiniMed with 2 filings, indicating moderate-to-high concentration among a few commercial players in retrieved records.
Patent Filing Distribution by Jurisdiction and Algorithm Cluster
Among 12 retrieved patent records in this dataset, WO filings account for the largest share of protective filings, reflecting global coverage strategies by commercial players. Algorithm cluster analysis reveals compound and MPC architectures as the most actively filed technology sub-domains in 2023.
Patent Filings by Jurisdiction — Closed-Loop Insulin Delivery (Dataset Snapshot)
WO filings represent the largest share of jurisdictions in this dataset with 10 records, followed by US (5), EP (3), CN (1), and IN (1), reflecting commercial assignees prioritizing international patent protection.
↗ Click bars to explorePatent Filings by Algorithm Technology Cluster — Closed-Loop Insulin Delivery (Dataset Snapshot)
Compound/hybrid algorithm architectures and MPC-based systems represent the most actively filed clusters in this dataset, with PID variants and learning-based controllers also represented across the 2009–2026 filing period.
↗ Click bars to exploreClinical Application Domains for Closed-Loop Insulin Delivery Algorithms
The retrieved dataset covers a broad spectrum of clinical application domains for closed-loop insulin delivery, ranging from pediatric and adolescent T1D management to inpatient critical care, neonatal intensive care, and pregnancy-specific algorithm adaptations.
Pregnancy & Obstetric T1D Care
Pregnancy represents a high-stakes domain requiring tighter glucose targets and rapidly changing insulin requirements. The Zone-MPC algorithm was tuned for pregnancy with daytime targets of 80–110 mg/dL in a 2022 study. The AiDAPT trial protocol (2022) and earlier 2011 studies collectively document a sustained MPC algorithm specialization for obstetric care across more than a decade of retrieved literature.
Specialty PopulationPediatric & Adolescent T1D Use
Pediatric-specific challenges including low insulin doses, unpredictable eating, and exercise variability are extensively documented in retrieved literature. A 2013 study demonstrated closed-loop therapy improvements in children under 7 years, and a 2014 feasibility study addressed overnight delivery in children aged 3–6 years using diluted insulin formulations. Adolescent home use (ages 12–18) was evaluated in a 2014 overnight study.
Pediatric PopulationInpatient & Critical Care Settings
A 2013 feasibility study evaluated fully automated MPC-directed therapy in 24 ICU patients over 48 hours. A 2019 study applied fully closed-loop insulin delivery to surgical and medical ward patients receiving enteral or parenteral nutritional support. A separate 2019 study documented personalized glucose targets using closed-loop delivery in end-of-life care settings.
Inpatient / ICUNeonatal Intensive Care Unit
A 2019 feasibility study applied closed-loop algorithms to preterm infants, and a 2018 study specifically addressed extremely preterm infants with birth weight under 1,200 g, a population with extreme glucose volatility. These studies represent the application of automated insulin delivery technology in its smallest and most fragile patient population documented in this dataset.
Neonatal PopulationKey Patent Assignees in Closed-Loop Insulin Delivery — Dataset Snapshot
In this dataset, Medtrum Technologies Inc. accounts for the highest filing volume with 8 WO/US records in 2021–2023, followed by Tandem Diabetes Care with 3 filings across US, EP, and WO jurisdictions. In retrieved records, commercial players in the US and China represent the most active assignees, with Harvard College representing academic IP activity.
Top Assignees by Filing Count — Closed-Loop Insulin Delivery (Dataset Snapshot)
↗ Click bars to exploreMedtrum Technologies Inc.
Medtrum Technologies is the most prolific patent filer in this dataset with 8 WO/US filings concentrated in 2021–2023. Their filings cover compound algorithm architectures simultaneously running rMPC and rPID controllers, dual-drug infusion systems for both hypoglycemic and anti-hypoglycemic agents, and bilaterally driven closed-loop pancreas mechanisms. Patent statuses in the dataset include WO and US filings with active and pending designations filed between 2021 and 2023.
China — CNTandem Diabetes Care Inc.
Tandem Diabetes Care holds 3 filings in this dataset across US, EP, and WO jurisdictions filed between 2021 and 2025. Their portfolio covers meal-announcement-based IOB tolerance adjustment (US, active, 2021) and an exercise-aware IOB modulation system with the most recent WO filing dated July 2025. The Control-IQ system based on Tandem’s technology achieved 94.2% median automation time across 9,451 real-world users in a 2021 study.
United StatesFive Emerging Signals in Closed-Loop Insulin Delivery Algorithms
Based on the most recent filings and publications in this dataset from 2022–2026, five directional signals are visible: fully closed-loop operation without meal announcements, LSTM and deep learning integration, exercise-aware IOB modulation, dual-hormone architectures, and wearable microneedle platform integration.
Fully Closed-Loop Operation Without Meal Announcements
A 2021 study reported 58% time-in-range for AndroidAPS operating without any meal announcements, signaling that algorithm developers are actively targeting the elimination of mandatory carbohydrate counting. This aligns with the stated future challenge identified in a 2022 review titled ‘Closed-Loop Insulin Delivery Systems: Past, Present, and Future Directions.’ Hybrid systems such as Control-IQ have already reached 94.2% automation time in 9,451 real-world users, setting a performance bar for fully closed-loop successors.
LSTM and Deep Learning Embedded in Closed-Loop Firmware
A February 2026 CN patent filing by Jiangsu University Affiliated Hospital (pending) directly embeds an LSTM blood glucose prediction model alongside a dynamic basal rate formula and least-squares metabolic parameter updating within the closed-loop algorithm. The filing targets 78% time-in-target and 0.8 hypoglycemic events per week, representing the clearest signal in this dataset of deep learning being operationalized within production-style closed-loop firmware. The 2021 in-silico study using double Q-learning with dilated recurrent neural networks validated on the FDA-accepted UVA/Padova simulator established precursors to this direction.
MPC vs. PID/rPID: Core Control Paradigm Comparison
Click any row to explore further.
| Dimension | Model Predictive Control (MPC) | PID / Risk-Transformed PID (rPID) |
|---|---|---|
| Core mechanism | Uses a mathematical model of glucose-insulin dynamics to predict future glucose trajectories and compute an optimal insulin dose over a receding horizon | Computes insulin dosing as a weighted sum of glucose error (proportional), accumulated error (integral), and rate of change (derivative) |
| Key commercial examples | Abbott Diabetes Care EP (2019, active); OmniPod personalized MPC (2018, ages 6–65); Diabeloop DBLG1 (ML-based, 2019 trial) | Medtronic MiniMed PID-IFB US (2014, active); Medtrum Technologies rPID WO (2023) |
| Specialty adaptation | Zone-MPC tuned for pregnancy (80–110 mg/dL daytime target, 2022 study); overnight MPC with sensor error model (Abbott, 2019) | IOB-corrected PID to prevent insulin stacking (2015 literature); dynamic setpoint adjustment (Medtronic, 2014/2018 US patents) |
| Hypoglycemia handling | Zone-MPC relaxes single-target objective to a glucose zone, reducing controller aggressiveness and hypoglycemia risk | rPID converts asymmetric blood glucose values from physical space to risk-symmetrized space, gaining precision while retaining PID simplicity |
| Real-world performance data | Control-IQ (MPC-based): 94.2% median automation time over 12 months, 9,451 real-world users (2021 study) | AndroidAPS (PID-based open source): 58% time-in-range without meal announcements in 2021 pig study |
| Patent filing activity (dataset) | 2 active patent filings in this dataset (Abbott EP 2019; Medtronic US 2018) | 2 patent filings in this dataset (Medtronic US 2014; Medtrum WO 2023) |
| Sensor error handling | Abbott EP (2019) embeds a stochastic sensor error model directly within the MPC optimization loop | Not explicitly addressed in PID/rPID filings retrieved in this dataset |
| Age range validated | OmniPod personalized MPC: age range 6–65 years (2018 study) | Pediatric-specific PID adaptations documented for children under 7 years (2013 study) and ages 3–6 (2014 study) |
Frequently Asked Questions: Closed-Loop Insulin Delivery Algorithms
PID computes insulin dosing as a weighted sum of glucose error, accumulated error, and rate of change. MPC uses a mathematical model of glucose-insulin dynamics to predict future glucose trajectories and compute an optimal insulin dose over a receding horizon. Zone-MPC further relaxes the single target to a glucose zone, reducing hypoglycemia risk. The rPID variant converts blood glucose values from physical space to a risk-symmetrized space to gain precision while retaining PID simplicity.
In this dataset, Medtrum Technologies Inc. is the most prolific filer with 8 WO/US filings in 2021–2023 covering compound rMPC/rPID architectures and dual-drug infusion systems. Tandem Diabetes Care holds 3 filings (2021–2025) covering meal announcement response and exercise-aware IOB modulation. Medtronic MiniMed and Harvard College each have 2 filings, and Abbott Diabetes Care has 1 active EP filing.
The February 2026 CN patent (pending) from Jiangsu University Affiliated Hospital embeds an LSTM blood glucose prediction model alongside a dynamic basal rate formula and least-squares metabolic parameter updating within the closed-loop algorithm. It targets 78% time-in-target and 0.8 hypoglycemic events per week, and represents the newest filing in this dataset and the clearest signal of deep learning being operationalized within production-style closed-loop firmware.
The Control-IQ system achieved 94.2% median automation time over 12 months across 9,451 real-world users in a 2021 study, with a mean user age of 42.6 years. The open-source AndroidAPS system achieved 58% time-in-range operating without meal announcements in a 2021 study. The open-source Loop system was evaluated prospectively in 558 users in a 2021 real-world study.
Retrieved literature documents algorithm adaptations for pregnant women with T1D (zone-MPC with 80–110 mg/dL daytime targets, 2022), children under 7 years and aged 3–6 years using diluted insulin formulations (2013–2014), adolescents aged 12–18 (2014), adults over 60 years evaluated with CamAPS FX (2022), 24 ICU patients over 48 hours (2013), surgical ward patients on nutritional support (2019), extremely preterm infants under 1,200 g birth weight (2018), and patients in end-of-life care (2019).
Among retrieved patents, WO filings account for the largest share with 10 records, followed by US (5), EP (3), CN (1), and IN (1). The predominance of WO filings reflects global protection strategies by commercial assignees. The appearance of a CN filing (2026) and an IN filing (2025) in the most recent period signals emerging geographic diversification of algorithm innovation into high-diabetes-prevalence Asian markets.
Data and insights on this page are based on a limited patent and literature dataset and are for reference only. Figures may not represent the complete technology landscape.