Pathway to Artificial Pancreas Systems Revisited: Moving Downstream

An electromechanical approach to improve glycemic control and quality of life for people with diabetes—an artificial pancreas (AP) (or an automated insulin-delivery system or bionic pancreas)—has been a long-sought technological goal of diabetes researchers (1). However, a number of significant challenges needed to be overcome to deliver an AP system to people with diabetes. The past 10 years have seen many of these challenges addressed, and recent studies have demonstrated compelling safety and efficacy of the prototype systems (2). Technical feasibility is only a step toward declaring victory. Research and development efforts will continue to improve upon first-generation AP systems. But, it is clear that resources will need to be deployed to address clinical adoption challenges—including device usability and reimbursement.

Research and development efforts over the past 10 years have addressed a number of the critical issues facing AP development, and recent studies again signal a move to a new phase and a call for a new road map to AP systems. A number of groups have demonstrated proof of concept with a variety of algorithmic approaches and closed-loop strategies and the data are compelling. In a Bench to Clinic narrative, Cefalu and Tamborlane (3) dive much deeper than this Perspective intends into the “how” of AP systems. They note quite astutely that “it is not the journey, but the destination that matters” (3).

It may come as a surprise to many clinicians, but a small and growing group of lay, “do-it-yourself,” technically savvy people with diabetes or loved ones of people with diabetes has been using semiautomated closed-loop systems at home for well over 2 years with impressive results (4–6). These systems, similar to systems being studied in academic studies, combine off-the-shelf insulin pumps and continuous glucose monitors with control algorithms (computer software that interprets glucose information and drives the dosing of insulin) powered by cell phone devices. The results from academia and these anecdotal reports are harbingers of a technological revolution in diabetes and signal that AP system availability is no longer a matter of “if” but rather a matter of “when.”

The ultimate metric for success of AP systems will be improved outcomes for people with diabetes. The technical feasibility demonstrated to date raises a series of very important questions that this Perspective will attempt to answer, or at least provide fodder for debate. AP systems will provide an interesting case study in the importance of a field having line of sight beyond the research laboratory and through to all of the stages of a commercial development pathway from regulatory approval, to reimbursement, to clinical adoption. For AP systems to improve diabetes outcomes, they will need to be designed to be impactful across a diverse group of people with diabetes and will need to be accessible. AP system accessibility will be driven by the value perceived by two other crucial stakeholders—health care providers and payers. The pathway to the development of an AP has become much more complex than the road map that was created in 2009, and a new road map that addresses postresearch considerations needs to be drawn.

This may seem like an obvious question, but the literature has demonstrated significant inconsistency in this definition and this inconsistency has led to confusion. This Perspective will focus on the nearest-term AP systems. The core elements of these AP systems will be an insulin infusion pump, a continuous glucose monitor, a control algorithm, and rapid-acting insulin analogs (in some cases, there may be a handheld control device). Reports in the literature use a wide variety of terminology (artificial pancreas, bionic pancreas, closed loop, automated insulin delivery device, and treat-to-range system [7–11]) because there is no, nor will there ever be, a singular AP. Rather, these systems will evolve over time to increase in automation, increase in sophistication, and increase in their ability to normalize blood glucose levels. In the near term, these “AP systems” will reduce hypoglycemia (low-glucose and predictive low-glucose management systems) through the reduction or cessation of insulin delivery, will begin to automatically dose insulin to target ranges (hybrid closed-loop systems, hyperglycemia/hypoglycemia-minimizing systems, and semiautomated insulin delivery systems), and eventually will dose hormones in addition to insulin such as glucagon and/or amylin (bionic pancreas, dual-hormone AP, and multihormone AP).

In 2006, JDRF launched an initiative intended to accelerate progress toward AP systems (12). At that time, many questions existed regarding the technical feasibility and safety of automated insulin delivery. These questions continue (13). A road map was published that intended to describe how an evolutionary process of system development could lead to the commercialization of clinically meaningful systems that addressed unmet needs in the management of diabetes (Fig. 1) (14). This road map intended to address these questions and, importantly, create clearly defined “target product profiles” that could guide research funding and commercial development based upon the state of technology at the time. This road map intended to shift the focus from replication of islet function with a machine to iterative improvements that addressed unmet clinical needs through increasing automation of insulin delivery.

It should be noted that this road map includes both systems that reduce/stop insulin delivery due to hypoglycemia or impending hypoglycemia (Fig. 1, boxes 1 and 2) as well as systems that automate the delivery of insulin (Fig. 1, boxes 3–6). All of these systems were described as AP systems, the criteria being the automation of the control of insulin delivery from an insulin pump. That said, there is certainly a significant step from reducing insulin delivery to increasing insulin delivery (Fig. 1, box 3), and many consider the first AP systems to begin at this point. This Perspective will discuss hypoglycemia-minimizing systems, but will use the term AP systems and synonyms to describe and discuss future systems that dose insulin and other hormones automatically.

The unmet medical need in diabetes is striking. Despite extensive knowledge of the damage of hyperglycemia and the passage of 23 years since the Diabetes Control and Complications Trial (DCCT) (15), glycemic control levels in the U.S. remain suboptimal. Current clinical evidence, including data from the T1D Exchange registry, paints a picture that is full of opportunity for significant improvement across all diabetes outcomes measured. The data on important diabetes outcomes paints a sobering picture.

In the U.S., A1C levels remain elevated with <20% of children and young adults and <40% of adults >25 years of age meeting A1C targets (16).

Hypoglycemia has been defined (17) and measured in a number of different ways. However, the data remain consistent that hypoglycemia is frequent by all measures and that it remains the limiting barrier to tight glycemic control (18–20).

Diabetic ketoacidosis is a life-threatening and costly acute complication. The rates remain elevated in the U.S. and globally (19,21).

Time in range is an intuitive metric for glycemic control that captures hyper-, hypo-, and normoglycemia in one simple view. This metric has only been possible to capture since the launch of continuous glucose monitoring (CGM) devices. A challenge in measuring time in target is the definition of the target range or the “normal” range. The definition of the range has varied in the literature (i.e., 70–105, 70–120, 70–180 mg/dL) (22–24). Whatever the target range that is used, people with diabetes are far from achieving normoglycemia.

The impact of diabetes remains significant on patients beyond suboptimal glycemic control. People with diabetes still suffer from significantly elevated levels of anxiety, depression, and other psychosocial issues due to a number of reasons, including the high burden that diabetes management places on the patient (25,26).

The obvious question is why? Why are glycemic control goals not being achieved? With crystal-clear evidence of the morbidity associated with hyperglycemia and hypoglycemia, why are only a fraction of people with diabetes achieving recommended glycemic and metabolic goals? There are many reasons that vary from individual to individual, but it is clear that the tools today do not easily allow for the normalization of glycemia for patients lacking β-cell function. Self-monitoring of blood glucose levels (27), continuous subcutaneous insulin infusion (28), CGM (18), and now low-glucose suspend (LGS) pumps (29) have all been demonstrated to significantly improve glycemic control either by reducing A1C or hypoglycemia levels. However, wide glucose excursions above and below the target range persist in almost all patients and attempts to achieve tighter and tighter glycemic control take more and more effort with diminishing returns. It is striking that children in the JDRF CGM trial, who were intensively managed, met with clinic staff regularly, wore CGM devices, mostly wore insulin infusion pumps, and finger-stick tested seven times a day, spent greater than 10 h a day with sugar levels above 180 mg/dL; and the adults spent more than 6 h a day with sugar levels above 180 mg/dL (18). It would be very difficult to argue that this was a noncompliant patient population. Clearly today’s tools have helped improve glucose control and in some cases reduce some of the burden of diabetes management. However, these data also clearly support that compliance is not the only barrier to optimal glycemic control.

There remains a debate in the literature and at diabetes conferences regarding the technical feasibility of AP systems given the state of the technology today. There are strong believers (14) and others who have questioned if a machine can ever replicate the sophistication of the islet (13). Unfortunately, much of this debate stems from a misframing of the argument. For example, those who believe that it will not be possible for today’s technologies to replicate the complex regulation of islet hormone secretions are probably correct. It is highly unlikely that we will normalize blood glucose levels though subcutaneous replacement of insulin alone (or even with insulin and glucagon). However a significant reduction of glycemic burden, both hyperglycemic and hypoglycemic, is possible with today’s technology, and as technology improves and insulin delivery and/or kinetics become more physiological, the bulk of glycemic excursions may be avoided. Therefore, we should not be comparing AP systems to physiological islet function; rather, we should be addressing unmet needs in current diabetes management that can be solved with technical solutions. Expectation setting will be very important as AP systems become commercialized and reach clinics. First-generation AP systems will not restore euglycemia and will not be fully automated, but they will significantly improve glycemia and reduce diabetes management burden in many patients.

AP systems are technically feasible today. Table 1 shows the six-step road map and references of representative studies demonstrating safety and efficacy. It is clear that AP systems outperform today’s standard of care significantly when benchmarked across a variety of diabetes outcome metrics. Larger studies are needed to gather A1C changes versus a randomized comparator, but the data to date is compelling and convincing. AP systems work.

A widely debated question in the field of AP research is whether the use of glucagon is necessary to build a safe system. Again, the data are convincing that insulin-alone–based systems will improve glycemic control, reduce hyperglycemia and/or hypoglycemia risk, and reduce some aspects of diabetes management burden (Table 1). The short answer to this question is therefore no—glucagon will not be essential for AP systems to reach the market. That said, results from insulin/glucagon studies have been outstanding and have generated enthusiasm in the field (9,30,31) and in the popular press (32). Conceptually, an insulin/glucagon approach is appealing. In the native/nondiabetic islet, the cross talk between α- and β-cells and the liver contributes to the intricate balance between gluconeogenesis and glycogenolysis and the maintenance of exquisitely tight glucose regulation within a very narrow range (33). A logical conclusion might be that AP systems would also benefit from such a bihormonal approach. However, more research will be needed to address significant questions regarding the consequences of glucagon infusion set failure and failure of the liver to respond to glucagon. Bakhtiani et al. (10) found that glucagon failed to prevent hypoglycemia and that these failures occurred more frequently when glucagon is delivered while glucose is falling rapidly, at a lower glucose threshold, and when there are high levels of insulin on board. El Youssef et al. (34) demonstrated that glucagon failed to prevent hypoglycemia ∼20% of the time in their initial studies. Pragmatic issues also remain, such as the need for a soluble pumpable glucagon and a dual-chambered pump and cost (35). In the coming years, it will be important to define the incremental value of glucagon and to define strategies to avoid glucagon “failure.” The Helmsley Charitable Trust, National Institute of Diabetes and Digestive and Kidney Diseases, JDRF, and the industry have invested significant resources to accelerate solutions to these challenges.

Another multihormone approach that has received less attention is the combination of insulin and amylin. Conceptually, this approach is appealing. The loss of β-cell function leads to the obvious loss of insulin production but also amylin production as well (36). Amylin plays an important role physiologically by suppressing glucagon production, contributing to regulation of gastric emptying, and impacting satiety. Amylin replacement through multiple daily injections of the amylin analog pramlintide has achieved limited uptake in the clinic (16). Pilot studies of multihormone AP systems using insulin and pramlintide have demonstrated impressive results (37). The same pragmatic issues exist for pramlintide as for glucagon, and this remains an area of continued investigation.

AP systems must reach people with diabetes and improve outcomes. To do so, they must receive U.S. Food and Drug Administration (FDA)/regulatory body approval, be commercialized, be reimbursable, and be adopted by providers and patients. In 2012, the FDA issued final guidance that provided a pathway for manufacturers to commercialize AP systems (38), opening the door for commercial development AP systems to reach the market. A major challenge to the translation of novel diabetes therapies into practice has been that the success of the therapy was judged with very narrow metrics that were heavily weighted to A1C. On the other hand, patients, clinicians, insurance companies, and government agencies weigh the benefits of new therapies by many other factors. Clearer descriptions of these factors may help inform the evolution and ultimate success of AP systems and future diabetes interventions.

Ultimately, success of a novel technology should be measured by the improvement in outcomes across the population with type 1 diabetes (T1D), which can only be achieved if clinically effective therapies are covered by payers, prescribed by physicians, and used by people with T1D who could benefit. Therapies well positioned for success will improve glycemic outcomes and reduce disease management burden at a cost consistent with the benefit provided by the therapy. In other words, therapies must provide a good value for people with diabetes but also for the health care professionals treating their diabetes and for the payers covering treatment, each of which has somewhat different perspectives on what constitutes value.

JDRF is developing a “T1D Scorecard,” a tool (or tools) that will provide a framework for evaluating the value of new diabetes technologies/therapies that is framed by the key attributes that are important to each stakeholder (Fig. 2). While Fig. 2 is not an exhaustive list, it highlights attributes that each stakeholder may weigh when evaluating new technologies. JDRF looks forward to working with other stakeholders to define a set of clinical outcomes by which to judge T1D therapies that is broader than A1C and to identify measures of disease management burden.

In light of the progress that has been made over the past 6 years, the six-step pathway for iteratively more advanced AP systems that was proposed in 2009 needs updating. This iterative pathway aimed to address technical limitations in glucose sensing and insulin delivery while positively impacting important diabetes outcomes. The data support that today’s technologies are ready to automate some degree of insulin delivery. In fact, for a small population of people, they already are. Boxes 1 and 2 in Fig. 1 are technically complete and the data support that they are safe and efficacious. Therefore, in 2015 the road map looks different and the obstacles have changed. Certainly, there are further technical advances that will allow for more sophisticated systems. But where should the focus be for that evolution?

Figure 3 proposes a new AP pathway—one that is bifurcated. Today, the pathway has evolved. Encouragingly, two of the six steps have been technically completed. In 2010, Medtronic released the first threshold-suspend system in Europe and in 2014 in the U.S. Such a system was a logical and important first step. Prior to its release and to date for all other insulin infusion pumps, insulin infusion continued even in the face of profound hypoglycemia and even while CGM alarms signal such lows. Cessation of insulin delivery during this time, while simple, held the potential to reduce hypoglycemic exposure and potentially profound hypoglycemia events. Studies of this system have validated this hypothesis (29,39). A more powerful approach to hypoglycemia reduction will be predictive suspension of insulin delivery with impending hypoglycemia. This approach improves glycemic outcomes (reduced hypoglycemia exposure, potentially reduced mean blood glucose vs. threshold-suspend system due to autoresumption of basal insulin delivery, and the interesting potential restoration of some counterregulation and hypoglycemia awareness) and provides reduction of burden (fewer alarms, reduced fear of hypoglycemia). It is also expected to be comparable in cost to threshold-suspend systems.

Whereas boxes 3–5 in Fig. 1 focus on reduction of hyperglycemia through an iterative increase in automation and box 6 focuses on multihormone approaches, today a more logical presentation of potential pathways is a split into two parallel avenues. One avenue will encompass automated insulin-alone delivery (AID) systems and another will be systems that incorporate another hormone or hormones.

Of these two pathways, AID systems are likely to reach the clinic first. Many studies have demonstrated safety and efficacy and the technical barriers are low. The first systems to reach the clinic will likely be “hybrid” treat-to-range or treat-to-target systems that require mealtime bolusing and then provide automated functionality that drives glucose levels back to a near-normal level (100–140 mg/dL in studies to date) during the rest of the day and night. The control algorithms that provide the automation of insulin delivery also provide the framework from moving seamlessly from systems that solely reduce/stop insulin delivery to those that add automated increases in insulin delivery as well (40). The focus of future AID system development will be across the three categories of the Diabetes Scorecard.

The main barrier to further improvements in glycemic outcomes (beyond a hypoglycemia hyperglycemic minimizer or hybrid AP systems) and to further automation of the system is the delay in the absorption and action of subcutaneously injected insulin. Other technological improvements, such as self-learning algorithms, integration of accelerometers, and better bolus calculators, may help incrementally improve glycemic outcomes and minimize user actions, but a more rapid insulin profile will be required to truly approach euglycemia and eliminate user prandial dosing in an AID system.

The more likely area of significant return on investment will be in the development of systems that reduce the burdens and barriers to CGM and insulin pump adoption. These will likely include calibration-free CGM, smaller CGM transmitters, smaller “tethered” and “patch” pumps, cell phone integration, and cloud-based data analytics.

The cost of AID systems should reflect the technology’s ability to impact glycemic outcomes and disease management burden. High-value systems will be appealing to all three key diabetes constituencies. Dual-hormone systems will likely require more time to reach the clinic (vs. AID systems) as practical and research questions are addressed. When compared with AID systems, a novel specialized pump will be necessary, and a specialized dual-lumen or modified infusion set as well as the additional hormone will be needed. These additional costs will need to be quantified and considered against the potential for improved glycemic control with a dual-hormone approach. To date, there exists very limited data comparing dual-hormone to an insulin-alone AP system (31). Further studies comparing best-in-class insulin-alone approaches to best-in-class multihormone systems should be a top priority to clearly define the pros and cons of each approach. Reduction of hypoglycemia, particularly severe hypoglycemia, represents a potentially significant benefit to glucagon/insulin AP approaches. While potentially more expensive, they may demonstrate additional value to the people with diabetes, health care professionals, and payers, and data supporting this argument will be very important.

For the past half century, AP technologies have been the “holy grail” of diabetes treatment. However, the sophisticated glucose regulation provided by the islet and the related metabolic physiology are difficult to replicate with a machine. The evolution of portable, small, easy-to-use, and efficacious insulin infusion pumps, continuous glucose monitors, and control algorithms over the past decade has allowed for proof-of-concept approaches—AP systems—that, while not perfect replications of islet biology, may provide significant value. The first automated insulin delivery systems that automatically reduce hypoglycemia exposure are already commercialized and are being used in clinical practice. AP systems that begin the dosing of some insulin automatically are expected in the 2017 time frame in the U.S., as reported from the J.P. Morgan Healthcare Conference in January 2015.

This progress has raised new questions and areas of focus. Success of AP systems will be defined by better diabetes outcomes. Better diabetes outcomes will include more than improved A1C (41). Broader understanding of the important glycemic outcomes in particular patient segments is needed. Furthermore, technologies will only be adopted and better outcomes achieved if they provide good value—improved glycemic outcomes and reduced burden at a cost consistent with the benefit provided by the technology. Ideally, next-generation diabetes therapeutics, AP systems, and beyond will improve glycemic outcomes, reduce burden, and provide value beyond today’s therapies and approaches. Improvement on any of these three scores for all of the three key stakeholders—patients, health care professionals, and payers—will portend well for novel technologies. Therapies that do not provide good value to all three stakeholders will face challenges to being widely adopted.

Acknowledgments. The author thanks his colleagues at JDRF, particularly Dr. Richard Insel, Cynthia Rice, Jessica Roth, Campbell Hutton, and Dr. Vincent Crabtree, for their thoughtful suggestions. He also thanks Dr. Roy Beck, John Lum, the JDRF Artificial Pancreas Consortium, and David Panzirer of the Helmsley Trust for countless hours studying, discussing, and debating the future of AP systems and thanks the diabetes device manufacturers who have supported AP research over the past 10 years and are working to deliver commercial AP systems to people with diabetes.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.