Future Direction of Artificial Pancreas Roy W. Beck, MD, PhD JAEB Center for Health Research Tampa, Florida
Financial Disclosures Dr. Beck does not have any personal conflicts of interest His employer, the JAEB Center for Health Research has received research or consulting funding or supplies for research studies from the following companies: Dexcom Abbott Diabetes Care Roche Animas Insulet Ascenia
OUTLINE History of Artificial Pancreas Development The Present: Medtronic MiniMed 670G System Near Future AP Systems Limitations of Current/Near Future AP Systems Future Directions Insulin delivery Glucose sensing Algorithms Summary/Take Away Messages
Biostator, circa 1980
Potential Pathway to an Artificial Pancreas Aaron Kowalski, circa 2008
Evolution of Medtronic Systems Medtronic Veo Pump: Low Glucose Threshold Suspend Medtronic 640G: Predictive Low Glucose Suspend Medtronic 670G: Hybrid Closed Loop
MiniMed 670G System FDA Approval Study (3 mos, N=123) Baseline Study Mean HbA1c 7.4% 6.9% Time in target 71-180 mg/dl 67% 75% Time <50 mg/dl 1.0% 0.6% Time <70 mg/dl 5.9% 3.3% Time >180 mg/dl 27% 25% Time >300 mg/dl 2.3% 1.7% Mean Glucose 150 mg/dl 151 mg/dl Sensor Wear 95% Time in Auto Mode 87% Bergenstal RM, et al. JAMA. 2016;316:1407-08. Garg SK, et al. DTT. 2017; 19: 155-73.
Near Future AP Systems
LBL014346 Rev 001 AP Systems in Development Using Dexcom CGM
Tandem Diabetes Care Predictive low glucose suspend algorithm integrated into t:slim X2 pump with Dexcom G5 CGM sensor Preliminary studies completed 90 subject pivotal outpt crossover trial starting imminently with FDA approval expected in 2018 Automated insulin delivery system in collaboration with TypeZero Pivotal study to be conducted as part of the International Diabetes Closed Loop (idcl) Trial led by the Univ of Virginia (Kovatchev, Anderson)
Tandem Predictive Low Glucose Suspend Algorithm CGM values are generated by the sensor every 5 min. Linear regression of the last 4 points (20 min) used to project future glucose value 30 min into future. If predicted glucose value <80 mg/dl or CGM value <70 mg/dl, system suspends insulin delivery. Insulin delivery resumes when CGM values rising or if insulin has been suspended for >120 min in last 150 min.
Tandem PLGS System
Tandem PLGS Feasibility Study ADA, 2017: Greg Forlenza-Barbara Davis Center 10 participants Mean age 22 yrs, T1D duration 13 yrs, HbA1c 7.4% 2 clinical sites (Barbara Davis Center and Stanford University). Outcomes Suspension of insulin delivery as intended per algorithm Median reference BG at insulin suspension = 88 mg/dl. 100% agreement between algorithm-recommended action and actual pump action. Resumption of insulin delivery as intended per algorithm Median reference BG at insulin resumption = 83 mg/dl. 100% agreement between algorithm-recommended resumption and actual pump resumption.
Tandem/TypeZero TypeZero is spinoff from the Univ of Virginia group Collaboration with Tandem using the UVA algorithm on Tandem pump with Dexcom sensor Pivotal study will be conducted as part of the idcl project funded by NIDDK
Omnipod Horizon TM Automated Glucose Control System Figure: Conceptual Rendering Omnipod pump Dexcom sensor Algorithm integrated in Pod
Omnipod Horizon TM Automated Glucose Control System 6 preliminary CRC studies with meal and exercise challenges 82 participants: adults, adolescents, and children 3384 hours of system use
Omnipod Horizon TM Hybrid Closed-Loop in Adolescents Mean glucose 153 mg/dl with HCL Time <70 mg/dl 60% lower with HCL compared with run-in phase
Omnipod Horizon TM Hybrid Closed-Loop Most recent adult study N=12 Mean glucose 136 mg/dl Time in range 70-180 mg/dl 85% Time <70 mg/dl: 1.5% Overnight time in range 94% and time < 70 mg/dl 0% Pivotal studies coming
Beta Bionics Boston University/MGH
Beacon Hill Study: February September 2013 Summer Camp Study: Summers of 2013 & 2014 Bionic Pancreas Multi-Center Study: June 2014 April 2015 Bionic Pancreas Insulin-Only Study: October 2015 January 2016 Bionic Pancreas Glucagon-Only Study: April June 2015 Bionic Pancreas Set-Point Study: August 2015 Dec 2016 Bionic Pancreas Monitoring Study: April May 2016
Bigfoot Biomedical System consists of an Asante pump body (disposable) married to durable controller (no screen or buttons) that talks to CGM and includes a control algorithm. Smartphone to serve as the window to the system and complete user interface. Per clinicaltrials.gov, initial study in CRC completed Dec 2016 Pivotal trial soon
Animas Animas pump and Dexcom CGM Hypoglycemia-hyperglycemia minimizer algorithm Preliminary studies completed AP program is temporarily on hold until strategic options for the company have been determined
Other Companies and Systems Cambridge MPC Algorithm (Roman Hovorka) Uses Medtronic 640G pump/enlite 3 sensor and Android phone Being tested in NIDDK-funded DAN05 trial DreaMed Diabetes (Moshe Phillip) Collaborating with Medtronic Fuzzy logic post-meal algorithm to be tested in NIDDK-funded FLAIR crossover trial
Other Companies and Systems Dialoop Cellnovo patch pump with Dexcom CGM Studies in progress Cellnovo Cellnovo path pump with UVA algorithm Studies pending Inreda Bihormonal pump, 2 CGM sensors for redundancy, fully automated algorithm In early testing Sensionics/Roche??
NIH/NIDDK Funded Projects NIDDK is funding 4 projects, all of which are coordinated by the Jaeb Center for Health Research idcl testing a control-to-range algorithm of Kovatchev et al 10 clinical sites in U.S., Netherlands, France, Italy DAN05 testing MPC algorithm of Hovorka et al 6 clinical sites in U.S., U.K. FLAIR testing fuzzy logic meal-time control algorithm of Phillip et al incorporated into Medtronic 670G platform 7 clinical sites in U.S., Israel, Slovenia, Germany BPTT testing MPC algorithm of Damiano/Russell/El-Khatib in insulin-only system and in biohormonal (insulin and glucagon) system 16 clinical sites in U.S.
Limitations of Current/Near Future AP Systems and Future Directions
Main Limitation of Current CL Systems Rapidity and longevity of insulin action Need for announcing meals and manual insulin bolusing Insulin catheters: approved for up to 3 days and glycemic control worsens with duration of catheter insertion
Insulin Kinetics
Faster Acting Insulin Aspart (orange)
Future for Insulin Delivery Need faster acting insulin action and shorter duration of effect to better approximate physiologic response of normal pancreas Need better catheters for subcutaneous insulin delivery Is there a limit to how rapid insulin effect can be with subcutaneous delivery? Alternative insulin delivery methods
Enhancing Subcutaneous Insulin Action Diasome s Hepatocyte Directed Vesicle Technology Additive to commercially available insulins Has been shown in nonclinical studies to deliver insulin directly and preferentially to hepatocytes Clinical studies: lower post-meal glucose concentrations compared with standard insulin administration More studies needed
Intraperitoneal Insulin Delivery
Insulin Levels with Peritoneal Insulin Delivery Versus Subcutaneous Nathan et al. Am J Med 1996
Post-Meal Rise with Subcutaneous Insulin Versus Peritoneal Delivery Nathan et al. Am J Med 1996
Intraperitoneal Insulin (orange)
Clinical Benefits of Intraperitoneal Insulin Delivery vs. SC Delivery Faster insulin action (on/off PK-PD) Tighter glycemic control Higher portal plasma insulin levels Lower systemic plasma insulin levels Lower incidence of hypoglycemic events Faster recovery from hypoglycemia Restoration of glucagon response Renard E, et. Implantable Insulin Pumps. A Position Statement about Their Clinical Use. Diabetes Metab 2007
Limitations of Implantable Pumps for Intraperitoneal Insulin Infusion Therapy Surgical procedure to implant and repair Risk of infection Catheter obstruction with fibrous tissue or omentumcausing under delivery of insulin Cost of device implantation and refilling pump reservoir
Eric Renard Observation Patients managed with an implantable pump and intraperitoneal insulin infusion do not want to go back to traditional insulin therapy
Patients with Implantable Pump & Intraperitoneal Insulin Delivery Life is easier with intraperitoneal insulin delivery Blood sugars and insulin action become far more predictable Fewer hypos If hypo occurs, it is far milder and recovery is faster Makes a profound difference in our lives We just plain feel better
CGM Limitations and Future Directions
CGM Accuracy Over Time Abbott Libre (12.3%) Medtronic Enlite 3 (10.5%) Faccinetti A. Sensors 2016
Dexcom Sensor Accuracy Through the Generations MARD 8% G6
Accuracy of Current CGMs by Sensor Day Mean Absolute Relative Difference Day 1 ~Day 3-4 ~Day 7 Dexcom G5 10.7% 8.0% 8.5% Medtronic Enlite 3 13.0% 8.9% 9.5% Abbott Libre* 13.8% 13.3% 12.3%
How Acccurate Does CGM Need to Be for a Closed Loop System Kovatchev et al Analyzed study data and conducted in silico simulations Simulated 7 CGM accuracy levels with MARD from 3%- 22% Concluded that using CGM for insulin dosing feasible and safe if MARD 10%. Below 10% not much gain unless it relates to a reduction in outliers
Factors Affecting CGM Accuracy Calibration error Compression (eg- lying on sensor when sleeping) producing false low glucose readings Interference from acetominophen producing false high glucose readings (current Dexcom sensor) Lag ~5 minutes Accuracy not as good during rapid change-- but can recognize this with alert arrows First day accuracy not as good as later accuracy
Future: Sensors Further improvements in accuracy, longer duration, less calibration, miniaturization Implantable, particularly peritoneal, can reduce the lag time and enhance concordance with blood glucose
Algorithm Limitations and Future Directions
Algorithm Limitations Subcutaneous insulins not fast enough for a fully automated system (meal announcements needed) Many factors affect blood glucose concentration levels Current algorithms focus on glucose and insulin levels
Future Directions Multi-hormone Glucagon Pramlintide Multiple sensor inputs Exercise Others Adaptive algorithms using deep learning (artificial intelligence) to personalize the algorithm for a patient
Additional Sensors for Future AP Systems Courtesy of Frank Doyle
Take Away Messages The last few years have had major advances in automating insulin delivery to reduce hyperglycemia and hypoglycemia The Medtronic 670G system is a landmark achievement in the advances made in the last 100 years in the management of type 1 diabetes. Despite this great advance, there are limitations. The 670G and other near-future systems function extremely well in reducing hypoglycemia during day and night and producing a high percentage of glucose levels in range overnight. Daytime is a challenge with meals, exercise, and other factors that affect blood glucose concentrations Current and near-future systems require meal announcement and bolus determination
Take Away Messages Future advancements needed in either faster subcutaneous insulins (although there may be a limit) or alternate insulin delivery methods (eg, peritioneal) CGM will improve further but current generation sensors are good enough for a well functioning AP system Future algorithm development needs to account for exercise and other factors that affect blood glucose concentrations Multi-hormone systems (insulin plus glucagon and/or pramlintide) are on the horizon
Thank you to JDRF and NIDDK for its funding of AP research and to FDA for recognizing the importance of AP systems for patients and for helping greatly to facilitate both the conduct of studies and the review process to bring these products to patients as quickly as possible