AP@home

Abstract

In the past years the development of an artificial pancreas (AP) has made great progress and many activities are ongoing in this area of research. The major step forward made in the last years was moving the evaluation of AP systems from highly controlled experimental conditions to daily life conditions at the home of patients with diabetes; this was also the aim of the European Union–funded AP@home project. Over a time period of 5 years a series of clinical studies were performed that culminated in 2 “final studies” during which an AP system was used by patients in their home environment for 2 or 3 months without supervision by a physician, living their normal lives. Two different versions of the AP system developed within this project were evaluated. A significant improvement in glycated hemoglobin was observed during closed-loop conditions despite the fact that during the control period the patients used the best currently available therapeutic option. In addition, a “single-port AP system” was developed within the project that combines continuous glucose monitoring and insulin infusion at a single tissue site. By using such a combined device the patients not only have to carry one less device around, the number of access points through the skin is also reduced from 2 to 1. In summary, close cooperation of 12 European partners, both academic centers and industry, enabled the development and evaluation of AP systems under daily life conditions. The next step is to develop these into products in cooperation with commercial partners.

The aim of this article is to summarize the outcome of the European AP@home project. During this project artificial pancreas (AP) systems were used in daily practice (“at home”) by patients with diabetes (www.apathome.eu). To meet the objectives of this project a consortium was established which consists of 12 partners from 7 different countries across Europe; this includes experts in the fields of information and communication technologies (ICT), algorithms, numerical simulation and validation, materials science, system integration, telemedicine, performance of clinical trials in diabetes technology and patient acceptance, and the development of single-port AP systems.¹ The project was funded by the European Union (EU) within the 7th Framework Programme (FP7).

The ultimate goal of the AP@home project conceived during its planning phase was to run a large multinational multicenter clinical study at the project’s end with AP systems developed within the framework of this project. In the end 2 ‘final’ studies were done in parallel. During these studies the feasibility and benefits of using AP systems under at home conditions for patients with diabetes was evaluated.

Background

To combine a continuous glucose monitoring (CGM) system to an insulin pump via a computer—which calculates the appropriate insulin infusion rates depending on the current glucose readings—is not a new idea; recent studies have confirmed that this indeed is possible.² Already some decades ago attempts have been made in Europe to develop an AP system.^3,4

One of the major bottlenecks for a successful implementation of an AP system in daily practice is that the currently available algorithms relating the glucose information to insulin infusion rates have to handle widely different insulin requirements during the day: from the more stable situation during the night toward the challenging situation after a carbohydrate rich meal. In other situations a reduction in insulin supply is required, for example, after physical exercise. In Europe a number of internationally renowned academic research groups intensively and successfully have worked on the optimization of such algorithms.^5-7

Until recently, different AP approaches (it is said >80 groups in academia and companies are working on this worldwide) were studied under highly controlled conditions in clinical research centers (CRC). The main reasons for doing so are safety concerns and regulatory requirements. During a recent workshop of the US regulatory agency their representatives highlighted this from their point of view:

• - The glucose measurement results provided by recently available CGM systems are not reliable enough to allow usage under daily life conditions.

• - The algorithms developed so far do not enable optimal coverage of insulin requirements after meals or during the night/after physical exercise.

• - The insulin pumps used in the different AP systems are not fail safe enough (all medical devices used in such studies were not developed and approved for this purpose).

• - The considerable variability of insulin absorption from the subcutaneous (sc) tissue after insulin administration further adds to the difficulties to achieve adequate glycemic control under all circumstances of daily life with the current AP systems/control algorithms.

It will not be before a better understanding of these factors is achieved—which will allow optimization of the closed-loop algorithms—that an AP system will become available that is truly suited for daily diabetes management of patients at home under all possible conditions.

In the past 40 years a frequent statement made in publications and presentations about AP systems was “in five years’ time such a system will be practically available.” Raising the hopes over and over again by saying that a technical “cure” of diabetes will become available within a relatively short period of time, without fulfilling this promise in reality, has induced a lot of frustration and skepticism at the patient and physician level. However, the recent developments in diabetes technology has brought CGM systems, insulin pumps, and small computer (in the format of smart phones) to a level that enables the successful development of a practically usable AP system.

Final Results of the AP@home project

As stated, aim of this project was the development of a wearable AP that automatically regulates insulin delivery in the home environment, enabling patients with diabetes to live a close to normal life. Two final AP@home studies investigated this: The “Hybrid Artificial Pancreas Study” was performed by 3 clinical research centers (Study 1; AMS, MPL, and PAD) and the “AP@home04 Study” was performed in 3 other clinical research centers (Study 2; CAM, GRZ, and PRO).^8,9

The purpose of Study 1 (“Hybrid Artificial Pancreas Study”; Hybrid-AP) was to assess the effectiveness of using an AP system for 2 months employing a model predictive control (MPC) and control to range (CTR) algorithm during the evening and night-time in the home environment in patients with type 1 diabetes, while these use their usual therapy (open-loop) during day-time (HYBRID). The study also aimed at demonstrating the feasibility of prolonged use of such an AP system on patient quality of life. The Hybrid-AP used an AP architecture based on the Diabetes Assistant (DiAs) platform developed by the University of Virginia, the Accu-Chek Spirit Combo Pump (Roche Diagnostics, Mannheim, Germany), and the Dexcom G4 CGM system (Dexcom, San Diego, CA, USA). These 3 components of the AP system were connected wirelessly. The closed-loop algorithm was developed by 1 of the consortium partners (PAV) and runs on an Android smartphone through the embedded DiAs platform. During a randomized cross-over open label study in patients with continuous subcutaneous insulin infusion (CSII) treated T1DM each patient underwent 2 periods of 2 months with a sensor-augmented pump (SAP) therapy. During 1 of these periods, called “HYBRID” period, the AP system was activated each day from dinner to wake-up time. Patients started either with the control phase (open loop, insulin pump with SAP feature) or with an AP system in the HYBRID mode, depending on the randomization. A total of 12 patients per center were recruited (36 in total). The duration of the study for each patient was up to 24 weeks (with 6 visits) with a recruitment period of up to 3 months. This study showed that evening-and-night closed-loop increased the proportion of time in target (glucose between 70 and 180 mg/dl) by 8.6% (95% confidence interval [CI] 5.8% to 11.4%, P < .0001 between 20:00 and 8:00; primary endpoint), reducing the time below target when glucose was less than 70 mg/dl by 1.6% (95% CI 1.0% to 2.3%, P < .001) and lowering glycated hemoglobin by 2.0 mmol/mol (95% CI 1.8 to 3.5 mmol/mol, P = .047). The conclusion was that 2 month closed-loop treatment at home under free-living circumstances is feasible, improves glucose control, and reduces the risk of hypoglycemic events. Closed loop from dinner until breakfast may be a safe first option for closed loop treatment.

The second final study (AP@Home04 study) aimed at evaluating whether day and night closed-loop insulin delivery for 12 weeks under free living conditions is superior to usage of real-time CGM in adults with type 1 diabetes and suboptimal glucose control on insulin pump therapy. The FlorenceD2A automated closed-loop system was used and a MPC algorithm was developed by 1 of the partners (CAM). The system comprises of FreeStyle Navigator® II (Abbott Diabetes Care, Alameda, CA, USA), Dana R Diabecare (Sooil Corp, Seoul, South Korea) subcutaneous insulin infusion pump, and a MPC-based glucose control algorithm running on an Android smartphone. Communication among the closed-loop system components was wireless. A total of 46 adults aged 18 years and older with type 1 diabetes (T1DM) on insulin pump therapy were recruited through diabetes clinics and other established methods. In total, 33 participants completed the study. This study showed that day-and-night closed-loop increased the proportion of time when glucose was between 70 and 180 mg/dl by 11.0% (95% CI 8.1% to 13.8%, P < .001; primary endpoint) while reducing mean glucose by 11 mg/dl (95% CI 6 to 17mg/dl, P < .001), reducing the area-under-the-curve when glucose was less than 63 mg/dl by 39% (95% CI 24% to 51%, P < .001) and lowering glycated hemoglobin by 4 mmol/mol (95% CI 1 to 6 mmol/mol, P = .002). The conclusion is that a 12-week application of closed-loop under free-living circumstances is feasible, improves glucose control, and reduces the risk of hypoglycemic events. A clinically relevant optimization of glucose control is achieved when closed-loop is applied day and night.

Components of an AP-System

Basically all AP systems under investigation consist of the same components, that is, a CGM system, an insulin pump, the insulin to be administered, and a “computer” with an algorithm calculating the amount of insulin needed by the subject. Initially the AP@home consortium assembled an AP system by using existing components (CGM system and insulin pump) and connecting these to a small notebook. The software needed to communicate with these devices and the control algorithm ran on this notebook. This prototype was built for research purposes for the first clinical trials and was optimized and further developed through the entire timeframe of the project, by replacing the cable connection by low energy Bluetooth connection and replacing the notebook by a smartphone that connects to the CGM system and the pump, runs the algorithm, and transfers the data generated to the cloud.

CGM Systems

A reliable and accurate glucose measurement is crucial for the successful performance of each and every AP system. The AP@home consortium performed a head-to-head comparison of all sensors on the market at the time. It had been shown that the accuracy of needle-type CGM systems can be poor, especially in the hypoglycemic range.¹⁰ As the project was started, there was no standardized procedure to assess the accuracy and reliability of CGM systems that are introduced to the market. Most often premarket assessments of CGM systems are done by means of the Clarke error grid analysis, which enables a clinical assessment of the correlation between CGM reported glucose values and reference blood glucose values. However, this approach provides only limited insight into the performance of CGM systems.

In an AP@home study 3 marketed CGM systems (Animas Vibe with Dexcom 4A sensor [Animas Corp, West Chester, PA, USA and Dexcom Inc, San Diego, CA, USA], Abbott Navigator I [Abbott Diabetes Care, Alameda, CA, USA] and the Medtronic Minimed Paradigm system with Enlite sensor [Medtronic, USA Minneapolis, Minnesota]) were studied in parallel in CRC and at home.¹¹ This was the first head-to-head comparison of these 3 CGM systems under intended use. In a comprehensive assessment, accuracy, assessed both under CRC and home conditions, and longevity also beyond manufacturer specified lifetime was determined. A difference was observed with the Navigator I and Enlite CGM systems outperforming the DexcomG4A system when assessed on day 1 of use at the CRC, while analysis of the home phase showed superior accuracy for Navigator I and DexcomG4A with a relative underperformance of the Enlite system: Overall mean absolute relative difference (MARD) (SD) measured at the CRC was 16.5% (14.3) for Navigator I and 16.4% (15.6) for the Enlite system, outperforming the DexcomG4A (20.5% [18.2]; P < .001). Overall MARD when assessed at home was 14.5% (16.7) for Navigator I and 16.5% (18.8) for DexcomG4A, outperforming the Enlite system (18.9% [23.6]; P = .006). One reason for this difference can be that CRC assessment of accuracy occurs during a relatively brief period of several hours with frequent sampling at predetermined times of reference values, encompassing the entire postprandial profile, whereas the home phase allows for accuracy assessment over several days. However, reference values were measured only a limited number of times per day and therefore cannot encompass the entire postprandial profile. It appears as if the Dexcom G4A system needs a longer warm-up period. This study also shows that sensor life can be extended beyond manufacturer specified lifetime by reactivating the sensors when the sensor session has ended, thus improving their cost-efficacy.

In conclusion, this trial showed that during CRC assessment differences between CGM systems could be observed, but these differences were different during the assessment at home. It is also of interest to note that the experimental approach used in this AP@home study was used again in a similar study evaluating more recent CGM system generations.¹²

Beside the analytical performance of the glucose sensor itself, the next step is to establish an appropriate signal processing of the provided data. One objective of AP@home was the reduction of noise in CGM signals and to reduce sensor delay, leading to the so-called smart sensor.¹³ Conventional filtering approaches used to reduce noise from CGM signals are of limited usage, also because they introduce significant intersensor, intersubject, and intrasubject variability to the signal-to-noise ratio (SNR). A novel filtering method developed by the AP@home consortium estimates and adapts the optimal filtering parameters in real time, on the basis of the current SNR.¹³ Subsequently, algorithms for on-line enhancement of the CGM signal were developed and tested. In fact, interstitial glucose (IG, measured by CGM) and blood glucose (BG, measurable infrequently by a laboratory method) were registered to exhibit differences which are due to the kinetics, which, roughly speaking, introduces a delay (metabolic part), and to sensor calibration (sensor part).¹³ To optimize sensor performance, 2 novel strategies were designed, exploiting CGM data plus 4-5 reference BG samples per day taken by fingerpricks, and assessed them on data collected by the clinical partners:¹³ The first strategy is based on an extended Kalman filtering (EKF) strategy.¹⁴ The second strategy is based on discrete deconvolution.¹⁵ Both are able to simultaneously perform calibration and time-continuous BG estimation and both exploit physiological information on plasma-to-interstitial glucose dynamics.

For the preprocessing of the CGM data the AP@home project developed an algorithm, working in real time, to improve the precision of CGM data. The algorithm is based on a statistical smoothing criterion, its parameters can be automatically updated in real time to adapt to possible variations of the signal-to-noise ratio occurring during a given monitoring, and can work in cascade to any CGM device.

In conclusion, application of software algorithms in real-time showed that CGM data can be smoothed by about 50% in all subjects, without introducing any significant delay.¹³ The 2 strategies described above, help to enhance the accuracy of CGM data. The first showed an enhancement of the accuracy of about 4 mg/dl with respect to original CGM data.¹⁴ The second strategy allowed an accuracy improvement of about 7 mg/dl with respect to original CGM data.¹⁵ While the first strategy has a burn-in time, which may become significant, the second strategy only requires the collection of a portion of CGM signals comprising a rising front, which may take 2-2.5 hours. In addition, the second strategy requires a lower number of self-monitoring fingerpricks than the first.

Insulin Pumps

When the AP@home project was started it was unclear if there is a difference between the conventional insulin pumps with an external insulin infusion set and more recently marketed patch pumps with no visible infusion set. A head-to-head comparison between the 2 types of pumps was performed.¹⁶ It was evaluated whether patch-pumps and conventional pumps showed differences in insulin absorption rates and reproducibility of absorption. In a multinational, open-label study serum insulin and blood glucose profiles were measured after bolus administration of prandial insulin via a patch-pump (OmniPod Insulin Pump [PP]) versus a conventional pump (Medtronic Paradigm Pump [CP]) in 20 patients with type 1 diabetes. The patients came to the CRC for 2 blocks of visits: each block consists of 2 visits while wearing the PP or the CP for 3 days.

Significant differences were only observed for intrapump changes, that is, not between the PP and CP. The intrapump differences include lower maximum glucose swing, lower average glucose levels, lower maximum glucose levels and lower area under the curve on day 3 of pump use versus day 1 of pump use. These findings suggest that the duration of use for a given insertion site is determining these differences.

Insulin Absorption

Insulin pumps deliver an insulin bolus to cover prandial insulin requirements. This bolus is usually administered as a pulse over a relatively short period of time; however, in practice the length of this period depends on the bolus size chosen and on the pump model used. Some pumps (eg, those manufactured by Animas) infuse 1 unit of insulin per second; that is, a typical bolus of 10-20 units of insulin is administered in less than 1 minute. Other pumps (eg, those manufactured by Medtronic or Insulet) infuse 1 unit of insulin in 40 seconds. In this case infusion of a similar bolus takes 7 to 14 minutes. Subsequently absorption of insulin administered is likely slower in the latter case and postprandial glycemic excursions would be higher.

To study the absorption kinetics associated with a short versus long bolus duration patients received an infusion of the same insulin dose (15 U Lispro, Eli Lilly) over 15 seconds (infused with an Animas IR2020 Insulin Pump) or 10 minutes (infused with a Medtronic Paradigm 512 Insulin Pump). Blood glucose was kept constant in euglycemic glucose clamps by means of an intravenous glucose infusion. Insulin infusion with a short duration showed significantly improved pharmacodynamic properties in comparison to long bolus duration in patients with type 1 diabetes:¹⁷ Earlier onset of insulin action (21.0 ± 2.5 vs 34.3 ± 2.7 minutes, P < .002) and shorter time to reach maximum insulin effect (98 ± 11 vs 125 ± 16 minutes; P < .005). This indicates that rapid insulin application may be able to better mimic the insulin kinetics seen in healthy persons after a meal. Because most AP approaches infuse insulin in short pulses depending on the current requirements, it should be taken care that at meals the required insulin dose is applied rapidly.

Algorithm

Many different algorithms for calculation of the appropriate insulin dose during closed-loops have been described; recent progress in the development of algorithms has focused on MPC algorithms.^1,2 This type of algorithm aims to address the challenging delays when insulin is infused subcutaneously and glucose changes are monitored in the subcutaneous tissue.^4,18 To evaluate the 2 different algorithm approaches available within the AP@home consortium, a head-to-head study was performed comparing the MPC algorithm developed by 1 of the AP@home consortium partners at the University of Cambridge (CAM)^5,6 and the algorithm developed by the partners at the Universities of Padua and Pavia (PAD; Italy) together with engineers in Virginia and Santa Barbara (USA).^7,8 The algorithms were implemented on different hardware platforms.

During the study days patients were in the 3 main life conditions (rest, meal, and exercise), that is, blood glucose was controlled for 23 hours either by 1 of the algorithms in closed loop (CL) or by patients themselves in open loop (OL), during 3 hospital admissions including standardized meals and exercise. This study was a multicenter, randomized, 3-way crossover, open label trial in 48 patients with type 1 diabetes mellitus for at least 6 months, treated with CSII.

To date, this is still the largest multicenter CL trial performed with overnight and daytime CL in the clinical research center. Interestingly enough both algorithms were quite comparable in their performance, that is, no significant differences could be observed in a series of parameters evaluated to characterize glucose control. However, with both algorithms time spent in the low glucose range was reduced in comparison to OL. Reduction of time spent in hypoglycemia is important in view of future home use of such algorithm driven insulin infusion systems. At the time the control algorithms were apparently tuned more conservatively, as they have shown to be able to lower mean glucose while preserving the reduction in time in hypoglycemia in later studies.

Telemonitoring

The aim of the AP@home project was to develop an AP system that operates safely in the patients’ home environment. Especially during the first trials it appeared to be important to stay in touch with the patients while using the AP system. This was done by means of a telemonitoring platform developed in this project.¹⁹ This platform collects parameters acquired through the AP system, along with other parameters concerning the functionality of the AP system itself, and forwarded those to a centralized Hospital Agent where all patients’ electronic health records were incrementally built and stored. Once data became available at the Hospital Agent, they were accessed by a dedicated application allowing the treating staff to monitor in almost real time the CL therapy.

Regulatory Aspects

To run clinical studies with a medical device requires approval by a local Ethical Committee and a governmental agency in most European countries. With the new Medical Device directive from March 2010, the barriers for approval of clinical studies with AP studies were raised considerably. When the first clinical studies were performed within the framework of the AP@home project it became obvious that there is quite some heterogeneity in the interpretation and handling of this directive by the regulatory agencies in the different European countries. Thus, getting approval for performance of AP studies is not only getting more difficult, it requires very different levels of intensity in different countries. This is one of the reasons why the final at home study of the AP@home project (see above and below) was split up into 2 separate trials performed in parallel. However, in Europe unsupervised usage of AP systems at the home of patients with diabetes eventually was possible. This is clearly needed to evaluate the true performance of a given AP system.

Steps From the Clinical Research Center to the Patients’ Home

At the time the AP@home project was started (and no AP system was available) the focus was on optimization of the CL algorithm. To have a high development speed the algorithms were tested and improved in simulators (in silico testing). After the optimization phase and the selection of appropriate hardware components (see above) the AP systems developed in our project were tested for 24 hours in the CRC under very close supervision by nurses, doctors and technicians. Based on this experience the first studies in the home environment of patients for 1 week usage of CL were initiated. After the successful performance of these first at home studies (Transition studies), the next step was the performance of a 2-month hybrid CL study under daily life conditions and a 3-month unsupervised full CL study at home, respectively.

Simulations of the Algorithm

Before simulations of the CL algorithms could be performed, 2 simulators were developed by members of the AP@home consortium to test the different algorithms. A deep in silico robust analysis has 2 goals: one is to analyze and compare the robustness of different algorithms and the other is to compare the capability of the 2 simulators to reproduce several difficult scenarios. The results have been obtained with about 96 000 hours of in silico experiments where 2 algorithms were tested on both simulators.

The principal components of a simulation environment include

1. a mathematical model of glucose regulation representing a virtual population with T1DM

2. the glucose measurement model

3. the insulin delivery model

A realistic representation of the virtual population with the appropriate level of inter- and intrasubject variability observed in vivo is considered key to reliable simulations. Moreover, a comprehensive validation is an essential requirement. Two distinct software packages—AP simulators—had been developed by AP@home partners (CAM and PAD/PAV). Both simulators were designed specifically to test glucose controllers and hence these simulation environments reflect the setup of a closed-loop clinical study. The simulators include a population of virtual subjects with T1DM necessary to represent intersubject variability and allow the user to define a variety of simulation scenarios.

The simulations showed that the 2 CL algorithms studied are quite robust with respect to important uncertainties and/or patient variability. Intensive in silico trials can help to improve the capability of the 2 simulators to reproduce real patient behavior but also to better tune the controller to enhance robustness against specific uncertainties.

24-Hour Closed-Loop at CRC

The second step was an intensive evaluation of the performance of the complete AP system with real patients under highly controlled conditions. The head-to-head comparison trial of the performance of the CAM and the PAV/PAD algorithm versus OL was to ensure that the performance was sufficient in the 3 main life conditions (rest, meal and exercise) over 24-hour periods.²⁰ This could be documented.

One Week at Home

Based on these encouraging results the next step was usage of the AP system by the patients in their home environment for 1 week. This was done in 2 different settings, with 2 different algorithms within 2 different regulatory frames. In 1 setting the iAP control algorithm was implemented on a suitable wearable platform (the DiAs). In addition, the AP@home telemonitoring approach was tested. Six patients with T1DM were enrolled at the University of Padova. A CGM system and an insulin pump were placed on the subject at the study admission. Both components were connected wirelessly to the DiAs platform through a Bluetooth hub implemented on a secondary phone. As a result the system overall performance was satisfactory. Risk mitigation procedures guaranteed safe conduct of the trial when the system was not properly functioning. The controller appeared to be properly tuned, improving consistently versus open-loop on dinner and night control. This 1 week at home study established the confidence that for the first time AP system are ready for larger clinical studies under daily life conditions. In a second multicenter transition trials 13 subjects were studied for 42 hours. Time in target was increased by 20% and time in hypoglycemia was reduced.²¹

In the other setting the performance of day and night CL based on the CAM algorithm was studied in the CRC (23 hours) and under free living home conditions (7 days).²² Also the results of this study demonstrate feasibility of unsupervised day and night CL glucose control under free living conditions in adults with type 1 diabetes. Usage of CL in relatively well controlled patients showed that compared to current best therapy, CL increased the time with glucose in the target range by a median 11% and reduced mean glucose levels by 0.8 mmol/l. Importantly, these improvements were achieved without increasing the risk of hypoglycemia while reducing total daily insulin usage. Furthermore, CL was able to maintain glucose in the target range in both day time and night times. Other benefits include reduced glucose variability and high glucose excursions.

The results of the above described studies paved the way for the 2 final AP@home studies, described above under the heading ‘Final Results of the AP@home project’

Next Steps

The development of AP systems is still in its infant stage. One important next step in the optimization of algorithms is their individualization.²³ The most promising approach is learning algorithms, which adapt to the patient better every day. The AP system will start with an algorithm that is suitable to bring the patients’ blood glucose levels safely into nearly normal range and by the learning capacity of the algorithm the mean blood glucose level (and subsequently the HbA1c) can be reduced further.

A critical challenge for AP algorithms is unannounced exercise. At least currently external parameters have to be fed into the CL algorithm to adapt insulin administration appropriately during exercise. One option is that an accelerometer provides information about the movement of the patient; alternatively recording of the heart rate can provide relevant information that can be integrated into the algorithm.

Another challenge is meals and especially the breakfast. If the CL algorithm only reacts on the rising blood glucose level caused by the meal, the infusion of the necessary amount of insulin will always start too late. With inappropriate circulating insulin levels postprandial blood glucose levels will increase into the hyperglycemic range after each meal. If the algorithm initiates infusion of too much insulin to reduce the hyperglycemic excursion, the applied insulin remains active beyond absorption of glucose from the meal in the gut. This increases the risk for late postprandial hypoglycemic episodes. Currently CL algorithms therefore need an announcement at least of large meals to function properly. An option to support the CL algorithm further is to not only announce the meal but also to indicate the size of the meal (small, medium, large). It could be even better if the carbohydrate content of the meal would be announced to the CL algorithm. Some CL algorithms need an estimation of the carbohydrate content by the user (eg, via a bolus calculator) to cover meals appropriately. Ideally the carbohydrate content of a given meal would be calculated by a smart image recognition software. The user would only have to take a picture of the meal. If the carbohydrate content is then fed automatically into the CL algorithm a further improvement in glucose control can be expected.

In addition to further improvements of CL algorithms the hardware components of the AP systems will undergo improvements. Currently the patient has to carry 2 medical devices (CGM system and insulin pump) plus the smart phone (along with a transmitter / battery packs). In the future the number of devices might be reduced to 1; this would make practical usage of the AP system more comfortable. In principle it is also possible to integrate the control algorithm and other parts of the CGM system into the pump. So in the end, the patient has to carry only 1 device with when using the AP system.

Single-Port System

A considerable hurdle for achieving patient acceptance for daily usage of any AP system is that a number of devices have to be carried around and handled all the time. The reluctance of many patients is already illustrated now by the large number not willing to use a CGM system regularly in addition to their pump in daily life. One possibility to improve the acceptance of an AP system is to reduce the number of skin perforations to 1 (“single port”); this requires combing the glucose sensing with the insulin delivery.^24-26 Such an approach can be assumed to improve patient comfort. Therefore, for a number of years such “single port” systems are under development, mainly in Europe.

One of the key questions with this approach is, has the infused insulin an impact on the glucose levels measured at the same site, for example, does the infused insulin induce a local drop in tissue glucose levels that results in a difficult to compensate deviation of the systemic glucose levels? Recent data from clinical-experimental studies (conducted by partner GRZ) show that the local interaction between infused insulin and glucose measurement via 2 needles inserted close to each other is negligible, that no relevant interference takes place.^24-26 Thus, a single-port approach is feasible, there is no need for more than 1 needle for both functions. By inserting a special indwelling catheter (ie, a microperfusion or a microdialysis probe) into sc adipose tissue of patients with diabetes and using the catheter for simultaneous insulin delivery and glucose sampling, the glucose concentration observed at the tissue site of insulin delivery correlates well with that seen in plasma.

During the AP@home project 2 different approaches were pursued in parallel to develop a single-port system:

• - Approach 1 relies on a nanoporous glucose-stimuli responsive material which changes its permeability for insulin with the patient’s glucose levels.

• - Approach 2 uses an insulin infusion catheter with an integrated glucose sensor that is inserted through a single skin perforation.

Approach 1 (brought forward by SEN and LAU) was based on a glucose-responsive membrane; the system consists of 3 modules (from the outlet to the inlet):

1. A glucose responsive composite nanoporous needle inserted subcutaneously and capable of dynamically changing its porosity as a function of glucose concentration in the subcutaneous environment.²⁷ The needle was closed with a pressure valve which allows bolus insulin injection.

2. A flow-meter capable of dynamically measuring changes in permeability of the outlet using tiny amounts of insulin solution (50-500 nl).

3. An off-the-shelf microliter-pump capable of delivering precise amounts of insulin solution.

The insulin solution flow dynamics was measured by the flow meter and controlled by the glucose responsive needle whereas the overall amount of insulin infused in the body was precisely controlled by the microliter-pump. The glucose responsive composite nanoporous needle consists of hollow fibers, with nanopores that were modified with glucose sensitive phenyl boronic acid based (PBA) hydrogel able to change its volume in the presence of glucose. Promising formulations of PBA hydrogel with a specific affinity for glucose at physiological conditions have been published.^28,29

Approach 2 (developed by GRZ and 4a) modifies an insulin infusion catheter to accommodate a continuous, off-the-shelf glucose sensor (from Dexcom Inc, San Diego, USA). It was shown in a proof-of-principle study that simultaneous insulin delivery and glucose sampling for a single-port AP is possible.³⁰ Recently it was also shown that with this approach it is possible to perform CL experiments successfully.³¹

Limitations

At the end of the AP@home project 2 large multinational multicenter clinical studies with AP systems developed within the framework of this project were successfully performed in parallel. During these studies the feasibility and benefits of using AP systems under at home conditions for patients with diabetes was evaluated. Both AP systems used were no commercial products and before a commercial launch of a true AP system several issues have to be addressed:

• Size: Both AP systems investigated in the final large studies consists of a CGM (sensor with the corresponding receiver), an insulin pump and a smartphone running the algorithm. For a wireless connection between smartphone and CGM a different relay device was developed and used in the studies. This additional device increases the size of the AP-system.

• Connectivity: Crucial for the AP systems is the (wireless) connection between smartphone and CGM and insulin pump otherwise the glucose values cannot be used by the algorithm or the calculated insulin infusion rate cannot be transferred to the insulin pump. A stable wireless connection can only be established on a short distance and therefore it is necessary to keep the smartphone close to the CGM and close to the insulin pump that means close to the patient at all times.

• Battery life time: All 3 components of the AP system are battery operated and therefore it is crucial to charge the CGM, the insulin pump and the smartphone whenever possible, preferably each night.

• Hypoglycemia: The use of the AP systems significantly reduces hypoglycemia. This is a large benefit for the patients, but the AP systems cannot avoid hypoglycemia completely. If the AP system predicts a hypoglycemic episode which cannot be avoided by the AP system an alarm is raised and the patient has to intervene.

• Acceptance: The AP systems used in the large final trials were well accepted by all patients. This is in part due to the fact that only patients were included in the trials that are willing to use the AP systems. This is not naturally because there are patients as well who are not willing to let an external technical device control their blood glucose.

• Compliance: As already mentioned the final AP@home studies were conducted with highly motivated patients. However, many patients do not care much about their diabetes or are not able to use the current CGM and/or pump devices. As both AP systems need regular calibrations with capillary BG values and the operation of a CGM, an insulin pump and a smartphone, only those patients will benefit from the AP system that have a positive attitude to medical devices and are trying to seriously manage their diabetes.

• Regulatory: Within the multinational AP@home project the study protocols of the different AP studies had to be submitted to ethics committees and regulatory bodies in 6 different European countries. It turns out that the requirements were different in different countries. This was 1 of the reasons why at the end 2 different final AP@home studies were performed.

Conclusions

By pooling and integrating European knowledge, technologies, expertise and capacities, the work performed by the AP@home project over a time-period of 5 years has brought the idea of a European based AP system toward a system that works successfully under daily life conditions.

Even if a practically usable system is available now, there is the need to convert this into a commercially available product and to optimize the system further. It could also be that other approaches, for example, use of bihormonal pump (infusion of insulin and glucagon) might enable development of AP systems that avoid the need for meal announcements. Also from a cost point of view, it can very well be that AP systems of different complexity might be required for different patient groups. This has been taken into account for clinical trials that are aiming to support the usage of AP systems in comparison to standard care.

In line with the statement about the evolving development of AP systems, it is also of importance to clarify and define the expectations. The first generations of AP systems will most probably not be a ‘cure’ of diabetes in that sense, that glycemic control of the patients with diabetes is brought in the same range as that of healthy subjects at all times. A first aim is to be better than current practice; this would mean that, for example, the risk of developing hypoglycemia is substantially reduced, that glucose level remains in the euglycemic range approximately 75% of the time, and so on. Setting expectations at a realistic level is also important to get regulatory approval for an AP system to be used in daily practice. The emphasis of regulatory agencies is more on safety aspects, whereas diabetologists and patients are more focused on efficacy.

The AP@home project has not only brought CL technology in Europe a major step forward, it also brought together many scientists from across Europe to work together at a high scientific level. In contrast to previous approaches, the AP@home project does not only raise expectations, it was able to prove the feasibility of using AP system(s) under daily life conditions. Cleary such a system has to be efficient, simple and safe; otherwise it will not become a viable product that can be disseminated to patients with diabetes.