After years of technical and regulatory issues, progress is now well underway in the great push toward automated insulin delivery (AID) systems. This review highlights current and future AIDs. Sometimes these AID systems that are designed to significantly reduce the burden of diabetes care and better simulate the normal pancreas are incorrectly referred to as a closed-loop or artificial pancreas, terms that refer to a system that fully controls glucose levels like a normal pancreas. “Hybrid closed loop” better describes these systems that still require the wearer to evaluate how a meal or snack will impact their glucose, choose the bolus dose required, and learn how to adjust the control settings to obtain optimal outcomes.
Glossary of Pump Lingo
- ACE – alternate controller enabled insulin pump
- AID – Automated Insulin Delivery or hybrid closed loop
- APS – Artificial pancreas system
- Basal – all day steady release of insulin
- BLE –Bluetooth Low Energy
- Bolus – a quick insulin infusion – carb boluses cover carbs; correction boluses lower high BGs
- Bolus Calculator (BC) – calculates bolus dose to recommend
- CGM – continuous glucose monitor
- COB –carbs on board or undigested carbs still raising the glucose
- Correction Target – the BG a correction bolus aims for
- DIY – do it yourself
- Duration of Insulin Action (DIA) – how long a bolus lowers the BG – used to calculate residual IOB or BOB activity
- FDA – US Food and Drug Administration
- HCL – hybrid closed loop requires manual boluses along with helful automation
- IOB or BOB – insulin on board or bolus on board still actively lowering the glucose from recent boluses
- iPump – interoperable pump, AKA ACE insulin pump
- MDI – multiple daily injections
- MPC – model predictive control algorithm
- PDA – personal digital assistant or dedicated phone
- PID – proportional integral derivative algorithm
- TIR – time in range, percentage of time your glucose is between 70 and 180 mg/dL (3.9 and 10 mmol)
- TDD – total daily dose – an average of all basal and bolus doses of insulin
AID systems combine three components: an insulin pump, a CGM, and a software control algorithm, and sometimes a dedicated PDA or smartphone app for remote operation. The algorithm or “brain” of the system oversees insulin doses that attempt to regulate glucose levels amidst the daily impact of meals, exercise, stress, medications, hormone levels, and other factors.
Benefits of AIDs:
- Better glucose levels,
- Less nighttime hypoglycemia,
- More time spent in the target glucose range of 70 to 180 mg/dL (3.9 to 10 mmol/L),
- Most importantly, less time spent below the range, and
- A better quality of life with fewer burdens.
Although greatly reduced, risks remain in complicated AID systems as the FDA balances these risks against the significant daily risks and problems already faced by people who manage diabetes without an AID. Annoyances sometimes arise in bringing about better glucose values due to excessive alarms and alerts, fingerstick requirements for older CGMs, and reduced control over settings and glucose targets. AID wearers may also gradually unlearn the details needed to manually manage their glucose levels as automation takes over.
Anatomy of an AID
The control algorithm commands the pump to deliver short varied intervals of insulin based on a variety of factors. Doses are designed to reach a glucose target pre-selected by the manufacturer or selected by the user in open source AIDs. Lower glucose target choices are expected to be available in upcoming AIDs.
An AID’s algorithm or “brain” may reside in the pump or pod, in a locked-down PDA, as an app in a cell phone, or in both the pump and a PDA or cell phone app that offers more discrete operation. AID systems reduce or stop basal delivery to minimize a projected low glucose level and increase insulin delivery in spurts over short intervals of time when the glucose is trending high.
Sometimes called “micro boluses,” these spurts are actually either a decrease in the basal rate or an increase when the glucose is projected to go high. Factors that control an increase in insulin delivery differ significantly between algorithms, such as the current TDD, the current basal rate, estimated fasting insulin level, and more obvious ones like recent and projected CGM readings. In some systems, the upper insulin delivery is limited by a maximum multiple of the basal rate, sort of a basal-constrained “corrective” block of insulin, while in others it is a combination of the factors above. When data is downloaded, insulin delivery appears as small boxes that vary in height between zero (no delivery) to the max multiple of the basal rate that the system allows.
Predicted glucose values are based on the current CGM glucose reading, its trend line, the expected rise from recent carb intake, and any glucose reduction expected from the residual insulin on board (IOB). IOB is determined by the time entry for the duration of insulin action setting (DIA) or insulin action time (IAT) becomes a critical setting because it determines the estimated IOB.
Short DIA times hide insulin stacking and become a serious threat when the buildup of insulin exceeds an AID’s ability to reduce insulin through suspension of the basal rate. Short DIA times blind the wearer and the AID to residual units of stacked active insulin and make it impossible to determine the exact number of grams of carb required to treat a particular low glucose.
Safety is improved by wearing an accurate CGM that more reliably detects and predicts changes in the glucose level, and by entry of an accurate DIA time that allows the algorithm to determine exactly how much residual insulin activity remains from prior deliveries of insulin. Besides reducing insulin stacking and hypoglycemic events, an accurate DIA time allows other pump settings that actually control glucose levels to be optimized.
JDRF and FDA Step Up to the Fast Pace of Diabetes Technology
Interest in AID devices is quite high. The market share in 2017 of $90 million is expected to grow to $279 million by 2024.1 While this is encouraging, advances in medical devices and software for treating diabetes often outpace insurance company policies that allow a pump or CGM to be replaced only every 4 to 5 years.
Older corporate policies left diabetes device wearers encumbered with outdated and less accurate equipment for months or years until the entire system can be replaced. If a particular component is replaced, it may not work with other older components or those of another manufacturer. In this case, the wearer could be shackled to one manufacturer when another offers improved technology.
To address this issue, the Juvenile Diabetes Research Foundation (JDRF) encouraged the FDA in August 2017 to consider an Open Protocols Initiative that would allow plug and play AID components to be more readily interchanged, giving users more choices and better access.
Interoperable and Interchangeable Devices
In March 2018, the FDA introduced the first plug and play framework by giving Dexcom’s G6 CGM the FDA’s first iCGM designation. This allows the G6 to share glucose data with any other device that receives an interoperable designation. To obtain an iCGM designation, a CGM must demonstrate that it has sufficient accuracy and can work interactively with any iPump or iAlgorithm through BLE communication.
Interoperability opportunities rose dramatically in 2010 when the 4th version of Bluetooth (named after King Harald “Bluetooth” Gormsson who unified warring factions in 10th century Denmark), called Low Energy, or BLE, was released. BLE enables secure and reliable communication between nearby devices over greater distances than the older versions with less demand on the battery.
In October 2018, the FDA next introduced an alternate controller enabled (ACE) infusion designation for insulin pumps, sometimes referred to as an iPump. The ACE iPump designation assures delivery accuracy, reliability, cyber security, and secure communication with compatible external devices and with insulin dosing software.
Today’s manufacturers can now integrate their technology or software with other companies who have an interoperable designation without having to resubmit clinical research data involving the already-proven devices. Any new iCGM can work with any iPump or iAlgorithm. For example, an older iPump can download software that displays glucose readings from a newly released iCGM or provides better glucose outcomes from an improved iAlgorithm.
Do NOT Take an SGLT-2 While Wearing an AID!
Most diabetes medications can be used with an AID. However, SGLT-2s like Jardience and Farxiga are simply too dangerous. SGLT-2 medications pass excess glucose into the urine. In this situation, an AID would normally increase insulin delivery to lower the elevated glucose, but the SGLT-2 fools the AID because the glucose stays relatively low even as the body’s deficit of insulin worsens.
Ketoacidosis occurs too frequently in Type 1 diabetes. This is why the FDA has not approved these meds for those with Type 1. Taking an SGLT-2 while wearing an AID greatly magnifies this serious risk, especially when an infusion set starts to leak insulin. Choose either AID or SGLT-2, never both.
The FDA has detailed the type of studies and data required to demonstrate the performance a CGM or algorithm requires for each designation to reduce regulatory requirements and costs. This regulatory move has been widely welcomed by companies with leading-edge technologies who can now more rapidly develop and deliver innovations to the diabetes community.
More importantly, as this integration develops, it will allow users yearning for the latest conveniences to select the devices they want to use and more rapidly integrate new technologies. These plug and play upgrades to new CGMs or control algorithms will lower costs and enable faster insurance coverage, compared to being restricted to old components for years.
The devices below have been at the forefront of interoperable plug and play devices for diabetes:
In March 2018, Dexcom quickly received the FDA’s first designation for an interoperable device with its G6 CGM or iCGM after passing strict accuracy requirements. The G6 sensor comes factory calibrated so fingersticks are not required. The G6 can also be calibrated as needed if a meter reading or two disagrees with the CGM’s reading. Readings can be shared with up to 10 people and users can also request a vocal report of their current glucose from Siri. Dexcom’s next-generation 14-day G7 sensor will be disposable, smaller, and be available at a reduced cost.
Senseonics and Medtronic are working on an iCGM classification for a current Eversense implanted CGM and a new Zeus CGM respectively. Agamatrix, Medtrum, and Sooil, among others are working on new iCGMs.
Tandem Diabetes—ACE iPump
On Valentine’s Day, 2019, Tandem’s t:slim became the first pump to achieve the FDA’s new ACE (alternate controller enabled) infusion pump or iPump designation, partly because of their early adoption of BLE communications. Other current and future pump companies are expected to gain this designation. Bigfoot and Sooil have BLE in their current pumps, Tandem has BLE in a pump in development, and Omnipod has BLE in their current pod and a future one, and Medtronic has two new pumps with BLE that are undergoing testing.
Tandem Diabetes—iController and First Fully Interoperable Closed Loop
On December 13, 2019, Tandem’s Control-IQ software became the first to achieve the FDA’s coveted “iController” status. This was the final leg required for a fully interoperable AID system where any part can be replaced with another part that has the FDA’s interoperable designation for true plug and play. Currently, only Dexcom’s G6 CGM, the Tandem t:slim X2 pump, and Tandem Control-IQ algorithm have an interoperable designation, although many other companies are working to join.
In September 2019, DreaMed Diabetes, an Israeli company that develops personalized diabetes management solutions received FDA approval for its Advisor Pro fuzzy logic/artificial intelligence algorithm to guide insulin dosing decisions by healthcare professionals. Work is underway with Medtronic to determine if this can be applied to AIDs. Medtronic is also improving its PID algorithm for the 780 pump, and has an agreement with Tidepool to use its Loop MPC algorithm in a 770 pump version as backup, once the FDA approves it. It remains unclear whether any of these algorithms will attain an iController status.
Tidepool, a 2013 Palo Alto diabetes data startup, is working with the FDA on approval of its a commercial version of the open-source Loop AID algorithm. Tidepool Loop is designed to work with commercial Bluetooth insulin pumps, CGMs, and an iPhone or Apple Watch. Tidepool’s bet is that the Loop open-source algorithm, currently in use by thousands of people with Type 1 diabetes, might speed FDA approval.
However, in May 2019, the FDA issued a warning about DIY Artificial Pancreas Systems (DIY APS) after someone experienced an accidental insulin overdose while using one. Accidents can happen with any of the current pumps and AID systems, but reports of failure to the FDA are sketchy and no regulatory body currently tracks relative failure rates between different systems.
Tidepool hired Loop project lead Pete Schwamb and Katie DiSimone, who wrote much of Loop’s online documentation, to assist in the production of a commercial version. Omnipod and Medtronic, who are simultaneously pursuing their own new control algorithms, have signed up to use the Tidepool Loop iAlgorithm, once approved, for their future iPod and Bluetooth-enabled iPump, respectively.
Dexcom acquired TypeZero Technologies’ inControl Diabetes Management Platform in August 2018. This algorithm was developed at the Univ. of Virginia for automatic regulation of insulin delivery by an insulin pump.
As shown in the Comparison Table, TypeZero’s inControl pump dosing algorithm is already used in several AID systems to reduce hypoglycemia and improve blood glucose levels. Tandem with its iPump and Dexcom with its iCGM and iAlgorithm appear likely to produce the first fully interoperable hybrid closed loop before the end of 2019. Other companies are actively working toward approval of their iAlgorithms by the FDA.
User and Clinician Training
Adequate training for these complex glucose control systems is sorely lacking. Many pump wearers are switched to AIDs with the same or similar settings used on their insulin pumps. Unfortunately, this results in many pumps currently in use containing serious setting errors(link to setting errors page) as demonstrated by the poor A1c results that are common among pump wearers. Setting errors generate erroneous basal and bolus doses and place those starting on an AID at a serious disadvantage.
Clinicians and users are often unaware of which setting or settings will favorably impact glucose control after starting on an AID. Which setting should they change and by how much when readings are erratic? Which settings no longer have relevance? To address setting errors and other concerns, clinician and user training and guidance are sorely needed. We recommend you visit our Pump Dose Guide for setting suggestions before starting an AID or when attempting to salvage a bad start. (Also quite helpful for those on MDI.)
Current Status of Commercial and DIY AID Systems
Until recently, the Medtronic 670G, approved by the FDA late in 2016, was the only commercial hybrid AID system. Appearing first on the market has been a blessing and a challenge, exemplifying the business admonition that it is often better to be a fast follower than the first one out the door. Restrictions designed for FDA approval and protection against lawsuits resulted in a system having a cacophony of alarms and alerts for users, limitations to user control, and excessive time spent out of AutoMode.
AutoMode suffers to some extent in having only one setting that the user can change to improve their glucose levels. Basal delivery is determined by the system depending on the TDD, estimated fasting insulin level, etc., but these are out of the user’s control. The user may select values for both the insulin action time (IAT or DIA) and the insulin:carb ratio. Users are sometimes advised to lower the IAT to reduce elevated glucose levels, but this creates “unexplained” hypoglycemia caused by the hidden insulin stacking that this generates. Furthermore, it doesn’t correspond to the meaning of insulin action time, ie, how long insulin works in the body to lower the glucose level after a bolus is taken. To safely reduce elevated glucose levels, the user can only lower their I:C number.
Alarm exhaustion has caused many to turn alarms off or abandon the use of AutoMode entirely in favor of sanity and sleep. This is unfortunate, since AutoMode operation benefits many users who start with appropriate pump settings), bolus before meals, and calibrate the CGM at bedtime to avoid one of the potential middle of the night awakenings.
AutoMode is intended to minimize episodes of the glucose going below 70 mg/dL (3.9 mmol/L). The fixed glucose target of 120 mg/dL (6.7 mmol/L) in AutoMode adds insurance against hypoglycemia. An exercise glucose target fixed at 150 mg/dL (8.3 mmol/L) is also available, although to have much effect, the target needs to be increased to 150 mg/dL at least 2 to 3 hours prior to any serious increase in activity with today’s insulins since a reduction in insulin dosing takes this long to have much impact. Carb intake, although reduced, may still be needed during exercise. For major increases or decreases in activity, AutoMode operates from the previous 6 days history, so it takes a few days to adjust to any new increase or decrease in activity.
Medtronic will update its 670G algorithm due to numerous user issues in favor of a newer version of its proportional-integral-derivative (PID) control, Tidepool Loop, and/or the DreaMed algorithm in their next-generation 770/780 Bluetooth insulin pumps projected to become available in the 1st and 2ndhalves of 2020. Reviews by Gary Scheiner and Will Dubois cover pluses and minuses of the 670G.
Tandem Diabetes Control-IQ
An early advocate for interchangeability, Tandem Diabetes integrated Tandem Device Updater software into their t:slim insulin pump in late 2016 to enable convenient software upgrades over the Internet without the need to replace the pump itself. Tandem’s full Control-IQ AID system is now available as the first interoperable control software. Tandem’s interoperability allows it to work with Dexcom’s current G6 iCGM and is also expected to work with Abbott’s pending Libre 2 iCGM. People who currently wear the t:slim X2 pump can download Control-IQ into their pump for free by signing into their account, requesting a prescription from their health care provider, and passing an online training session in the system’s features and use.
Similar to DIY systems, Control-IQ provides users with far more control over their glucose. Basal rates, I:C ratios, and correction factors can all be adjusted to improve control. The DIA time is preset to 5 hours for insulin action to minimize insulin stacking as glucose levels are normalized. This is far safer and saner than systems that allow short DIA time entries that do not match the physiologic action of current pump insulins in the body.
The Control-IQ algorithm improves time in target by aiming to keep daytime glucose values between 112.5 and 160 mg/dL. Basal delivery is reduced if the glucose is projected to go below 112.5 mg/dL (6.3 mmol/L) to lessen hypoglycemia, while insulin micro-bursts are increased once the glucose is projected to go above 160 mg/dL (8.9 mmol/L). If the glucose is projected to go above 180 mg/dL (10 mmol/L), small correction boluses, equal to 60% of the expected correction bolus with a target of 110 mg/dL (6.1 mmol/L), are added each hour on top of the micro-bursts, raising the importance of having an accurate CorrF or ISF.
For sleep hours, a lower target range of 110 120 mg/dL can be set by the user. Experienced users who know how to effectively control their glucose sometimes select 23-hour sleep periods for their “day”. For exercise, Control-IQ has an option to keep the glucose between 140 and 160 mg/dL (7.8 and 8.9 mmol/L). This exercise option adjusts the glucose more quickly than the 670G for major changes in activity, such as an adjustment for the weekend warrior or starting or stopping marathon training. An alternate approach for exercise is to enter a duplicate user profile (easily done in this pump), and then lowering the basal rates and raising carb and correction factors in this profile. Activation of this profile allows rapid accommodation for heavy periods of activity or exercise.
Premeal boluses are required. However, if a meal bolus is forgotten, Control-IQ detects the rising glucose and automatically delivers about 60% of the usual meal bolus dose using a glucose target of 112.5 mg/dL (6.3 mmol/L).
In a pivotal research study, 168 participants using the Tandem Control-IQ system spent 2.6 more hours a day in the 70-180 mg/dL (3.9-10 mmol/L) target range and lowered their average A1c by 0.3% from 7.4% to 7.1% compared to a control group with the same pump and CGM but no automation. Everyone completed the study with 92% of the full six-month time period spent active in AID and 71% of this time spent in time in range. On a 5-point scale, participants rated the ease of use at 4.7, usefulness at 4.6, trust at 4.5, and desire to continue its use at an impressive 4.8 for several reasons: simplicity, no fingersticks with the Dexcom G6 CGM, near-continuous automated operation, and a minimal number of alarms. Clinicians conducting the Phase 3 trial reported that none of the people using this system wanted to discontinue its use.
Insulet Omnipod Dash
Following closely after Tandem’s t:slim X2, Insulet became the 2nd company to receive FDA’s ACE designation in September 2019, for its Omnipod Dash system. Dash meets the FDA’s cybersecurity, reliable communication, design, and transparency requirements. This ensures it can reliably and securely communicate with other BLE connected devices, as well as AID software, and that it can receive, execute, and confirm commands from these devices.
Dash is expected to communicate with Dexcom’s G6 iCGM and eventually with the Freestyle Libre 2 CGM that is expected to receive iCGM status. Insulet is working on two hybrid closed-loop algorithms, its own internal Horizon MPC algorithm and a partnership with Tidepool Loop’s algorithm that may be closer to FDA approval. Both are expected to have smartphone app controls.
A Loop DIY version for Omnipod Eros pods is already available.
DIY AID Systems
Although the FDA `does not officially approve them, open-source DIY AID systems are permitted to operate due to their non-profit status. At last count, over 1,500 people were using DIY systems. With no federal oversight, these systems do not undergo traditional clinical trial processes and standards, bypassing the slower paths required of corporations for FDA approval. As mentioned, one case of over-delivery of insulin was reported to the FDA in May 2019.
In theory, DIY systems may enable better A1c results with fewer highs and lows and more time in the desired range than the Medtronic or Tandem because they allow more customization of glucose targets and settings by the user.
The DIY movement pioneered interoperability. Dana Lewis, Scott Leibrand, and Ben West started the movement in 2014. They used Linux to get the system operational on a small Raspberry Pi computer plus a communication stick that used radio communication to talk with older Medtronic pumps and direct BLE communication to other devices. Other motivated and organized individuals and parents quickly joined this open source effort. Their current system uses a dedicated circuit board or an app on some Android phones that communicate with some Dana insulin pumps and Omnipods, as well as Dexcom and other CGMs.(#WeAreNotWaiting)
OpenAPS offers a “carbs required” alert when the projected decrease in basal rates is not sufficient to prevent a low glucose. Originally offered in the Deltec Cozmo pump in the early 2000s, this option should be mandatory in all AID systems. Unlike Loop and AndroidAPS below, meal boluses are delivered through the pump bolus calculator rather than a smart phone. Also available is Autotune and Autosens(https://openaps.readthedocs.io/en/latest/docs/Customize-Iterate/autotune.html) that transfer the day’s data at night into a Nightscount account to suggest changes in pump settings for the next day.
Loop DIY systems are based on an open-source algorithm that utilizes Riley Link hardware ($150) with a communication board and rechargeable battery in a plastic case. Additional components include a Dexcom CGM and an older Medtronic pump (See Figure 1)
Like all DIY systems, a substantial degree of tech-savvy is needed to personalize the pump, phone, and software. Steps include becoming an Apple developer (free or $99 per year), using Xcode software from Apple for programming (free), and potentially encountering programming errors for those first trying Xcode. For programming and setting change issues, a large online user manual details each step, and online chat groups and blogs provide lots of assistance.
Basal and bolus settings, BG targets, and carbs on board (carb digestion that is still raising the glucose) are adjusted through a Loop app on an iPhone 6 or later. No correction bolus is automatically given when the glucose is elevated, but Loop compensates by allowing the user to set its maximum basal rate at high values. This enables larger “correction bursts” over short time intervals. The user can carefully adjust their max basal rate setting upward to more aggressively correct elevated glucose readings once they’ve entered an accurate DIA time to avoid insulin stacking. Recommended settings for the maximum basal increase are 2 to 5 times the current basal setting.(ref) The Loop system tracks carbs on board (COB) and allows the user to set digestion times for fast, medium, and slow carbs when consumed, adding value to simple carb counting.
The Nightscout open-source data-sharing project through Heroku enables access to an Autotune option for advice on improving glucose readings. Autotune looks back at several weeks of Nightscout data to suggest changes to the basal rates, CorrF/ISF, and CarbF/ICR. Setting changes are limited to 20 to 30% above or below the underlying pump values, so starting with appropriate settings speeds progress. Commercial companies are expected to integrate some version of an Autotune concept into future AIDs. Percentage Profile Switch lets the user adjust both basal rate and ISF for quick adaptations for rapid control shifts for illness, activity changes, etc.
Developed by Milos Kozak and Adrian Tappe in Europe, the AndroidAPS system works on Android phones with current pumps with BLE such as Dana R and RS and Roche Accu-chek Combo and Insight. Again, Nightscout through Heroku enables an Autotune option for advice on improving glucose readings.
Math-based approaches have been developed by engineers to handle complex situations in multiple arenas where predicting outcomes using current inputs leads to better results, such as with today’s commercial airliners and self-driving cars. Diabetes is a perfect example of an arena where this can be helpful. The two most commonly used algorithms are model predictive control (MPC) that is considered more proactive or predictive, and proportional-integral-derivative (PID) that is considered more reactive. Many varieties of each are in development. A recent review of research studies on control algorithms found that “MPC performed as well or better than PID in all metrics.” MPC algorithms are currently used in most AIDs.
Model Predictive Control (MPC) algorithms are currently a favored design for advanced control systems in diabetes because their flexible design allows the system to handle multiple inputs through the use of quadratic equations. Exercise, as well as pregnancy in women, tends to be easier to account for with MPC. Each input can then be optimized in a constrained way every time a prediction is made, such as every 5 minutes when new glucose arrives from the CGM.2 Insulin delivery is determined by it ability to reduce a forecast glucose level against the desired target glucose levels over the next 1.5 to 4 hours. Researchers have developed numerous variants of MPC for use in AIDs.3
Safety constraints were recommended in one patent for MPC algorithms used in diabetes.4 These include a maximum basal rate increase for reduction of glucose levels that is no larger than 2 to 5 times the current basal rate, full basal suspension when the glucose is predicted to go below 77 mg/dL (4.3 mmol/L), reduced basal delivery any time the glucose is falling rapidly, and delivering only the current basal rate when an infusion set occlusion is inferred to more rapidly identify the real issue without introducing false insulin delivery data into the algorithm that creates turbulence.
PID algorithms adjust insulin delivery by the rate of change in glucose (derivative), its deviation from a target glucose (proportional), and the area under the curve between the measured and target glucose (integral). PID responds to the current glucose rather than to a predicted glucose, such as predicted affects from meals or activity. Other algorithms, such as those based on fuzzy logic are also being tested.
Many other companies— Insulet, Bigfoot Biomedical, Beta Bionics, Lilly Diabetes, and Roche — are in various stages of development and testing for AID systems. In addition, Diabeloop, a French company, has received the CE mark for distribution its AID in Europe.
All of these systems are insulin-only systems with the exception of Beta Bionics iLet they have taken the lead on a dual hormone system using insulin to lower the glucose and glucagon to raise the glucose and more effectively guard against hypoglycemia. Zealand Pharma has developed dasiglucagon that is now in clinical trials with the Beta Bionics iLet pump, while Xeris is working on a similar trial with a glucagon solution for bihormonal AIDs at Oregon Health and Science University.
Overcoming Inherently “Slow” Insulin
A major obstacle for all AID systems involves the inherently slow action times of current insulins compared to the far faster digestion times of most foods. Duration of insulin action of pancreatic insulin is less than 40 minutes. Infusion of commercial insulin below the skin is slower to start and has a much longer duration of action of at least 5 hours in both children and adults compared to natural insulin released by the pancreas that goes directly into the portal vein to the liver. Subcutaneous delivery of Novolog, Humalog, and Apidra takes 20 to 40 minutes before the insulin begins to lower the glucose.
Help appears on the way with the expected release of Lilly’s “ultra rapid” Lispro (URLi) insulin. Compared to Humalog (lispro), URLi reduces glucose levels by 28 mg/dL in Type 1 subjects and by 12 mg/dL in Type 2 subjects at one hour and by 31 mg/dL in Type 1 subjects and by 17 mg/dL in Type 2 subjects two hours after a test meal. URLi starts lowering the glucose after 21 min compared to 28 min for Humalog. The overall impact appears good. In the graphic on the left in the figure below, at 50 minutes after an injection, an elevated glucose level is lowered by 120 mg/dL with URLi (red line) compared to only 65 mg/dL with Humalog (blue line).
URLi appears to be more stable in insulin pump reservoirs than Novo’s FiAsp that appears to lose potency on about day two. Keep in mind that the DIA for URLi is unchanged from Humalog – it works as long as shown by the graph on the right but with an earlier peak in activity. A new formula for this faster decline in action will be needed for use in pumps and AIDs with this insulin.
Lilly’s patent describes URLi’s components as lispro insulin at concentrations from 40 to 500 U/mL, combined with citrate for faster absorption, and low concentrations of treprostinil (a prescription drug with potent vasodilator effects) that also enhances absorption. Citrate has to be balanced: more citrate speeds absorption but unfortunately also destabilizes the insulin. Treprostinil is balanced as well to allow local dilation of blood vessels without any systemic effects. Other more standard insulin ingredients include zinc, magnesium chloride, m-cresol, and glycerol.
Hybrid closed loops or AIDs are definitely a step forward and new systems will greatly reduce user burden. Advances are underway in all areas. The AID Comparison Table reviews current and future systems.
|Comparison of Current and Future AID Systems|
|Medtronic 670G||Guardian 3 CGM, PID algorithm||180u or 300u reservoir, 2-4 calibrations a day, no remote monitoring, I:C ratio is only control option, not capable of software upgrade, excess alarms for some users, too many exits out of AutoMode, BG target: 120.||CE: 2012
FDA: Nov, 2016, release in summer, 2017
|Medtronic 770G||Zeus CGM similar to Guardian 3, Tidepool commercial Loop MPC algorithm on iPhone, Riley Link.||BLE, easy software upgrades, calibrations unclear, multiple control options, remote monitoring, BG target selected by user. Loop usually works with BLE CGMs like Dexcom – unclear if this will occur.||FDA: pump and iCGM 1st half 2020?|
|Medtronic 780G||Synergy disposable CGM, updated PID or DreaMed Advisor Pro fuzzy logic algorithm.||BLE, easy software upgrades, no calibrations, user control options?, remote monitoring, BG target: 100 to 120 (150 for exercise), applying for iCGM designation, 7-day infusion set.||FDA: iCGM and AID 2ndhalf 2020 or 2021
FDA approved DreaMed Advisor Pro in Sept 2019
|Tandem t:slim||Dexcom G6 or Abbott Libre 2 iCGM,
TypeZero inControl MPC algorithm,locked-down PDA or phone app
|Color touchscreen, no calibrations, multiple control options, remote monitoring, always stays in AID, few alarms, auto basal shut off for predicted lows, basal increase above 160 mg/dL, hourly auto 60% of expected correction bolus minus IOB above 180 mg/dL, 300u reservoir||FDA: iPump ACE approval Feb 2019
FDA: Control-IQ now available and shipping
|Tandem t:sport||Dexcom G6 or G7 iCGM, TypeZero inControl MPC algorithm, locked-down PDA or phone app||Color touchscreen PDA, no calibrations, multiple control options, integrated bolus button, patch or pocket/belt wear, easy software upgrades. Target: 112.5-160 mg/dL daytime, 112.5-120 mg/dL overnight, operates independently of PDA, 200u reservoir||FDA: 2ndhalf of 2020 or early 2021?|
|Loop DIY||Dexcom G5 or G6 CGM, Old Medtronic pump or current Omnipod, OpenAPS, RileyLink BLE and radio relay box, Loop MPC app on iPhone or iWatch||Xcode software, fully customizable, alerts only for good reason, no calibrations with G6, multiple control options, remote monitoring, 180 or 300u reservoir, requires technical skill, BG target set by the user, Autotune||Non-FDA, available now: $150 for Loop, $99 for Apple developer license.|
|OpenAPS DIY||Dexcom G5 or G6 CGM, old Medtronic pump, Linux on a Raspberry Pi microcomputer, Open APS MPC algorithm||Linux software, fully customizable, alerts only for good reason, no calibrations with G6, multiple control options, remote monitoring, 2 or more profiles, Autosens settings guide, 300u reservoir, requires technical skill, BG target set by the user||Non-FDA, available now.|
|AndroidAPS DIY||Dexcom G5 or G6 CGM, BLE-enabled Dana R or RS pump or Roche Combo or Insight pump, AndroidAPS MPC software, Ruffy app, and LineageOS or Android 8.1 on phone||Android Studio software, fully customizable, alerts only for good reason, no calibrations with G6, multiple control options, remote monitoring, 2 or more profiles, Percentage Profile Switch lets the user adjust both basal rate and ISF for quick adaptation, 300u reservoir, requires technical skill, BG target set by the user||Non-FDA, available now.|
|Insulet Omnipod Loop||Dexcom G6 or G7 iCGM, Tidepool Loop app on iPhone, Tidepool Loop MPC algorithm||Color touchscreen PDA or app on select Samsung phones, pod can operate independent of PDA, no calibrations, multiple control options, remote monitoring, 200u reservoir, BG target set by the user, (39mm x 52mm x 14.5mm)||FDA: iPump ACE approval Sept 2019
FDA: Full system 1s thalf 2020?
|Insulet Omnipod Horizon||Eros pod, DASH, locked-down color Sansum Android PDA, Dexcom G6 or G7, U of Virg.TypeZero inControl MPC algorithm||DASH locked-down color touchscreen Samsung phone, easy software upgrades, no calibrations, remote monitoring, multiple control options, 200u reservoir, pod can operate independent of PDA, BG target: ?||FDA: iPump approval Sept 2019
FDA: Full system 2nd half 2020?
|Bigfoot Biomedical SmartLoop||Loop (Asante Snap) BLE insulin pump, Freestyle Libre 2 CGM, Tidepool Loop or perhaps TypeZero inControl MPC algorithm, phone app||Low starting price via monthly subscription, easy software upgrades, remote monitoring, 300u pen fill easy cartridge change, no calibrations?, multiple control options, BG target: ? Bigfoot was first to receive FDA’s fast track approval for its AID in November 2017.||FDA: 2019 or 2020
with Dexcom’s iCGM and iAlgorithm
ACE designation for Snap iPump 1sthalf 2020?
|Beta Bionics iLet Bionic Pancreas||iLet bi-hormonal pump, Dexcom G6 or Senseonics CGM, Gen 4 Touchscreen PDA, PID algorithm||No or few calibrations, multiple control options, remote monitoring, BG target: 120 mg/dL for insulin, 110 mg/dL for insulin/glucagon||FDA: 2020 for insulin only,
2021 for insulin + glucagon
|Roche Diagnostic Ltd.
|Accu-Chek Insight pump or Solo patch pump? DexCom G6 iCGM or Sensonics Eversense CGM, TypeZero Technologies inControl MPC algorithm
|2019 International Diabetes Foundation Closed-Loop research study involved 43 subjects at 14 sites in the US and Europe||CE: 2020?
Roche remains quiet in this area
|Diabeloop DGLB1 System||Wearable Kaleido patch pump with short infusion set, locked-down Sony Xperia Z1 PDA, Dexcom G6 CGM||No calibrations, multiple control options, remote monitoring, 200u reservoir, BG target: 100 to 130 mg/dL||CE: 2019
|Lilly Diabetes||Lilly Deka (Dean Kamen) 2” diam disk pump, Dexcom G6 or G7, McGill Univ. algorithm, dedicated controller or phone app, 3 ml reservoir||Reusable and disposable components, patch or pocket/belt wear, no calibrations, multiple control options, remote monitoring, use any luer lock infusion set, 200u reservoir, BG target: ?||FDA: 2020 or 2021?|
|SFC Fluidics||Hydraulic pump with 2-way safety valve and Percusense CGM housed in a single on-body pod, Diabeloop machine-learning algorithm, JDRF funded||Disposable patch pump?||FDA: 2020 or 2021?|
|Agamatrix||Agamatrix iPump and Subsidiary Waveform Technologies’ iCGM, Oregon Health & Science Univ. MPC algorithm, Apple or Android phone app
|One hour warm-up, glucose reports each minute, 14-day sensor wear, calibrations?, user options?, BG target: ? mg/dL||FDA approval late 2019 for Waveform iCGM?
2020 or 2021 for iPump?
|Medtrum||P6 EasyPatch disposable pump, S7 EasySense CGM, A6 TouchCare System with predictive low BG Suspend||Limited usage in Europe, rechargeable touchscreen PDA for pump control, colored “brain” part of pod is reused, 200u reservoir in white section is reused, lower cost monthly rental, integrated 5 mm 30 gauge stainless steel needle, 3-day wear||CE: 2019
FDA: 2020 for pump? Unclear if S7 CGM has accuracy for iCGM status
|Sooil Dana||Dana Diabecare RS BLE pump, OpenAPS MPC algorithm, Dexcom G5 or G6 CGM, Apple or Android phone||300u reservoir, BG target set by the user, full user calibrations, multiple control options, remote monitoring, only one proprietary reverse-luer-lock Teflon infusion set.||FDA: 2020?|
|EOFlow EOPancreas||EOPatch pump (9.9mm x 32.4mm x 12.9mm); EOCloud algorithm derived from early Type Zero software? Locked-down Android phone or Android app.||Disposable, CGM in a pod under development, 200u reservoir, integrated 30 gauge stainless steel needle, calibrations likely, multiple control options, remote monitoring, 3-day wear. EOFlow received fast-track approval from the FDA for its AID in March 2019.||FDA: pump 2020? AID 2021? Unclear if CGM has accuracy for iCGM status|
Suggestions to Improve AID Systems
- Train HCPs and users in the critical details required to succeed with each of these complex glucose management systems.
- Each algorithm should clearly state how each pump setting impacts the user’s glucose outcomes and how different settings interact. Will someone on a low basal rate be handicapped if insulin bursts are limited to a multiple of the current rate? Which settings or what circumstances determine the maximum correction dose? What benefit or danger arises from using a short or long duration of insulin action (DIA)?
- Allow two or more sets of settings: one for normal pump operation during CGM changes, etc., another with more aggressive settings for use during AID operation, another for team sports during the school year, etc.
- Provide an alarm for infusion set failure since this continues to be the most vulnerable component in AID systems.
- Track Carbs on Board with the user able to set duration of carb action times for fast, medium, and slow carbs (Loop and AndroidAPS have this).
- If the AID is unable to reduce the basal rate sufficiently to prevent hypoglycemia, the system should inform the user exactly how many grams of carb they need to prevent the pending hypoglycemia (the Deltec Cozmo insulin pump did this over a decade ago).
- Include the user’s weight as a setting for accurate estimates of the insulin to carb ratio and determination of the exact number of carbs needed to treat each low or pending low glucose. An accurate DIA is also required for acuracy.
- Allow the user or clinician to individualize the glucose targets once hypoglycemia risk is documented to be minimal.
- Once 14 days of glucose data are available, the AID system should provide common solutions for any pattern of hypo or hyperglycemia.
- Through human factor studies, confirm that a new user can take everything out of the delivery box, select initial personalized settings, and safely start on the AID on their own.
- Integrate activity monitors for rapid adjustments in insulin delivery for seasonal sports or a new sport or a change in work activities.
Written by John Walsh, PA, CDTC, and Ruth Roberts, MA
1 Coherent Market Insights. “Global Artificial Pancreas Device System (APDS) Market to Surpass US$ 279.2 Million by 2024.” heraldkeeper, 01 January 2019, http://heraldkeeper.com/market/global-artificial-pancreas-device-system-apds-market-surpass-us-279-2-million-2024-218824.html
2 B. Wayne Bequette, Ph.D. Algorithms for a Closed-Loop Artificial Pancreas: The Case for Model Predictive Control.
3 The Artificial Pancreas: Current Situation and Future Directions. Edited by Sánchez-Peña RS and Cherñavvsky DR.
4 Wilinska ME, Budiman ES, Hayter GA, Taub MB, and Hovorka R. Integrated closed-loop medication delivery with error model and safety check. US Patent 9,402,953 B, 2016.