Refining Insulin on Board with netIOB for Automated Insulin Delivery

Abstract

Automated insulin delivery (AID) systems enhance glucose management by lowering mean glucose level, reducing hyperglycemia, and minimizing hypoglycemia. One feature of most AID systems is that they allow the user to view “insulin on board” (IOB) to help confirm a recent bolus and limit insulin stacking. This metric, along with viewing glucose concentrations from a continuous glucose monitoring system, helps the user understand bolus insulin action and the future “threat” of hypoglycemia. However, the current presentation of IOB in AID systems can be misleading, as it does not reflect true insulin action or automatic, dynamic insulin adjustments. This commentary examines the evolution of IOB from a bolus-specific metric to its contemporary use in AID systems, highlighting its limitations in capturing real-time insulin modulation during varying physiological states.

Introduction

In individuals without diabetes, insulin secretion into the hepatic portal vein is dynamically regulated minute-by-minute by a complex array of neuroendocrine and nutrient factors that are “sensed” throughout the body.^1,2 Intricate feedforward mechanisms, predominantly neuroendocrine, alongside feedback processes which mainly involve glucose, fatty acids, and certain amino acids, orchestrate beta-cell insulin secretion. After being secreted into the portal circulation, the half-life of insulin is approximately three to five minutes, with ~80% of the insulin cleared during its first passage through the liver. ³ This physiological nuance of endogenous insulin secretion and action are unattainable in individuals with type 1 diabetes (T1D) managed with subcutaneous insulin, despite progressive strides in insulin formulations, glucose sensing technologies, and automated insulin delivery (AID) systems. ² Rightfully so, patients are educated that the time-action profile of prandial insulin is about three to six hours, depending on the type of insulin administered. Insulin should ideally be taken well before a meal to allow for its absorption and distribution to the various insulin target tissues that facilitate glucose disposal. Unfortunately, the slow onset of exogenous insulin action and slow decay still make glucose management challenging around meals and exercise.

The use of AID systems has helped individuals with T1D maintain glucose concentrations within a clinically acceptable target range (70-180 mg/dL) for large portions of the day.^4-6 Current commercially available AID systems rely on user-initiated carbohydrate and exercise announcements and readings from a continuous glucose monitoring (CGM) system inserted subcutaneously in the periphery to inform insulin infusion rates by the insulin pump. Yet, these algorithm-driven AID systems have inherent limitations. Notably, insulin is administered subcutaneously rather than delivered into the portal circulation, and its action profile spans up to seven hours, in stark contrast to the rapid biological half-life of endogenous insulin, which lasts about three to five minutes.^7,8 Moreover, modern AID systems often lack substantial feedforward inputs, limited primarily by the user-input of carbohydrate consumption, and occasionally activity features, with scant considerations for broader neuroendocrine fluctuations such as stress, illness, changes in menstrual cycles, and/or changes to whole body insulin sensitivity and/or insulin clearance rates.^9,10 These factors, which are often perplexing and unpredictable to persons living with T1D, still present significant challenges for the self-management of diabetes using current AID systems.

While continued advancements in diabetes-related technologies and therapies are anticipated, it is essential to consider the interactions between a user and their AID system. Also known as hybrid closed-loop (HCL) systems, AID systems currently require user-initiated actions (ie, administering meal boluses, adjusting basal insulin delivery, or setting temporary glucose targets) under varying physiological conditions, including exercise, meals, and stress. ¹¹ While numerous factors influence a user’s decision-making, including past glucose responses, current glucose levels and trends, and carbohydrate intake, one crucial piece of informative data for the user is “insulin on board” (IOB). However, we argue that the current representation of IOB in AID systems may be misleading, potentially affecting important insulin-delivery decisions.

What Is IOB and Why Do so Many Terms Exist to Try to Explain Insulin Action?

An understanding of IOB, and its role in managing T1D, necessitates an appreciation of both its historical origins and current applications. To date, a variety of terms have been used to elucidate the critical role of insulin within the body (refer to Table 1 for a glossary of terms). To our knowledge, this concept was first introduced under the term “bolus on board” (BOB) to help explain how bolus insulin action could remain physiologically “active” in its effects on circulating glucose concentrations (or glucose turnover) for up to seven hours, albeit with diminishing effects over time.^12-14 The concept of BOB was particularly valuable during patient education to help prevent “insulin stacking” that would sometimes occur when glucose concentrations rose following a prandial bolus, as patients were educated on the prolonged glucose-lowering effects of a bolus insulin dose. ¹³ Eventually, the acronym BOB was replaced with IOB, which, with the advent of insulin pump technology, is now widely recognized and used, but certain factors warrant consideration.

Table 1.

Glossary of Terms Related to Insulin Delivery Needs When Using a Standard Insulin Pump (Open-Loop) or Automated Insulin Delivery System (AID). Contribution of Each Term to IOB Is Also Noted.

Term	Description
Insulin on board (IOB)	Traditionally defined as the bolus or prandial insulin action within the body following bolus insulin administration, sometimes referred to as “Bolus on Board” or “Active Insulin.” This term is used to convey the concept that the duration of insulin action for any rapid or ultra-rapid insulin analog lasts for several hours after subcutaneous administration, depending on the amount and type given. The term “IOB” is appropriate for multiple daily injections (MDI) therapy and open-loop (standard) insulin delivery pumps, but can be confusing when used in the context of AID systems since its definition varies across systems (see below). The estimation of IOB depends on the amount of insulin taken and the duration of insulin action. In all AID systems, IOB includes both manually entered and algorithm-derived correction boluses (for projected glucose levels above the target range), but does not include basal delivery unless it exceeds the pre-programmed basal rate setting. The way AID systems estimate and report IOB to the user lacks consistency among various commercial manufacturers.
Basal insulin a	The amount of insulin required to maintain glycemia within a certain range in an unfed and non-exercising state, often termed as basal insulin needs in MDI and open-loop and AID insulin delivery systems. It is sometimes referred to as basal insulin “needs” or “requirements.” For MDI therapies and open-loop insulin delivery pumps, this form of insulin (or insulin delivery rate) is not considered a part of IOB. However, in AID systems, since insulin delivery rates may vary from the set basal insulin needs during open-loop mode, this term may not be as useful (or may be confusing) when describing insulin delivery rates.
User-initiated meal bolus	The act of manually administering insulin for a meal or snack, often related to estimating carbohydrate intake and other meal features such as fats, protein, and carbohydrate type. Users may adjust this insulin dose to mitigate the risk of hypo- and hyperglycemia based on factors like activity levels, stress, or illness. This manual bolusing is typically factored into the estimation of IOB.
User-initiated correction bolus	The act of manually administering insulin to address an observed increase in glucose level. This is usually calculated by considering the current glucose concentration, the target glucose level, and the user’s insulin sensitivity factor. Users may adjust this insulin dose in situations where they believe their insulin sensitivity is changing, such as during illness (when more insulin may be needed for correction) or after exercise (when less insulin may be needed for correction). This manual correction bolus is typically factored into the estimation of IOB.
Automated correction bolus b	The algorithm-driven administration of insulin in bolus form by an AID system when glucose is forecasted to exceed the treatment target range. Typically factored into the estimation of IOB by most AID systems.
Negative insulin on board	A concept term used to describe any reduction in basal insulin delivery compared to the standard pre-programmed basal rate in open-loop mode, applicable in both open-loop and closed-loop AID modes. Open-loop example: a 50% reduction in basal rate for 90 minutes during prolonged physical activity, where the usual rate is 0.8 units per hour at rest, would result in an estimated negative IOB of 0.8 U/h x 0.5 (for % reduction) x 1.5 hours = −0.6 U. Closed-loop example: setting an AID system to a higher glucose target leads to a 50% reduction in basal insulin delivery overall for 120 minutes of physical activity, compared to the usual basal rate delivery in open-loop mode of 1.0 units per hour. The estimated negative IOB is 1.0 U/h x 0.5 (for algorithm-determined percent reduction) x 2.0 hours = −1 U.
Net insulin on board (netIOB)	A newly proposed term aimed at encapsulating the overall active insulin effect, encompassing recent bolus insulin deliveries, whether manual or automated corrections, alongside any adjustments to basal insulin delivery from the pre-programmed rate (ie, the insulin delivery rate set for a specific time of day in open-loop). netIOB can be either positive or negative, representing a relative measure against the user’s baseline or basal insulin state at any given time of day. In essence, this metric reflects all changes (increases or decreases) in insulin delivery relative to the basal (unstimulated) state.

Several different types of control algorithms have been developed for AID systems, including model predictive control (MPC), proportional integral derivative (PID), and fuzzy logic (FL) controllers. Each of these systems have different ways of determining the patient’s insulin needs and they also differ slightly on how the insulin delivery occurs. As such, some of the terms used in this table are not applicable to all systems. ⁹ The traditional concepts of “basal” insulin and “bolus” insulin with the use of AID systems can cause confusion for users since both ways of insulin delivery are used to mitigate hypoglycemia and hyperglycemia and manage glycemia after food ingestion. The following terms (1) user-initiated and (2) algorithm modulated insulin delivery may be more appropriate in some cases. ⁹

^aMost AID systems require that basal insulin needs are “pre-populated” within the system so that the user can use their pump in an “open-loop” mode, such as in settings where the CGM sensor signal is unavailable or if they prefer not to use the algorithm-derived insulin delivery options (such as during exercise). Also note that the basal insulin delivery rate is also used by most AID systems as part of the algorithm-determined insulin delivery needs. If the algorithm-derived insulin delivery rate is above the usual open-loop rate, then this insulin should also be considered as IOB. If the insulin delivery rate is below the standard set rate, it is unclear how this metric might influence the AID’s reporting of IOB.

^bSome AID algorithms use a “reverse” correction if the glucose concentration at the time of bolus is below target.

All insulin pumps use rapid-acting insulin for both basal and bolus insulin needs and are programmed with a scheduled basal insulin infusion rate, meant to reflect resting physiologic needs. ¹⁵ Standard (i.e., open-loop) insulin pumps require user-initiated correction or meal boluses. To mitigate insulin stacking during meals, in 2002, the first bolus calculator was introduced in the Deltec Cozmo insulin pump. ¹⁶ This allowed the standard pump to display the amount of active insulin (ie, IOB) from recent mealtime and/or correction bolus doses, enabling users to consider this information before administering a subsequent bolus.^17,18 While standard pump users can issue temporary basal rates or activate different settings profiles, these changes are not reflected in IOB. It should still be noted, however, that commercial standard insulin pumps may use different modeling to calculate IOB, and the formulations are not readily shared. ¹⁹ Understanding IOB within the context of standard insulin pumps is relatively straightforward; however, the advent of AID systems complicates the concept of IOB. In summary, IOB may be thought of as “active insulin”, but it may not be a true reflection of physiologic insulin action. Insulin action does not peak at the time of administration, and several factors are known to influence the insulin absorption and clearance rate ¹³ (Figure 1). Moreover, the visual display of IOB on an AID system depends heavily on the duration of insulin action setting, which can be customized on some AID systems from anywhere between two and six hours. This setting can profoundly influence the “aggressiveness” of insulin delivery. ²⁰ As shown in Figure 1, if the duration of insulin action is set to three hours on an AID system for an individual using insulin lispro, the user may not realize that the peak insulin action of a bolus may occur closer to 1.5 to two hours after administration. ²¹ Moreover, after three hours, the individual may glance at the AID and see zero IOB, not realizing that the effectiveness of the last bolus is still at ~55% of the maximal effectiveness. Setting the duration of insulin action to five hours may allow the IOB estimate to be more reflective of the insulin action profile for insulin lispro, but this setting will place some constraints on the insulin delivery in an AID device after any bolus.

Figure 1.

A theoretical visualization of insulin on board (IOB) compared with the percentage of peak insulin action over a six-hour time span. The solid lines represent theoretical insulin on board for a five-hour (blue) and three-hour (purple) user-inputted insulin action time. The dotted orange line represents the percentage of peak insulin action for a typical rapid-acting prandial insulin (see text for further information).

AID systems use historic and/or current sensor glucose values and trends to predict glucose levels and then increase or decrease insulin delivery relative to the pre-programmed basal rate, to change the predicted glucose outcome. ⁹ This means there are a myriad of insulin adjustments per day, rather than a few user-initiated boluses and basal modulations, and the majority of these will be driven by the AID system that considers the time of insulin action and/or the IOB. ²⁰ As previously articulated in the literature, the traditional concepts of “basal” and “bolus” are meaningless for insulin delivered in an automated fashion, as most AID systems use a combination of basal and bolus insulin delivery to mitigate hypoglycemia and hyperglycemia. ⁹ However, due to a lack of transparent reporting to the user, understanding this cumulative insulin delivery in AID systems is challenging and raises numerous questions.

How Some AID Systems Appear to Use and Misrepresent IOB to Their Users

The advantage of AID systems lies in their capability to automatically increase, or decrease, insulin delivery in response to changes in glucose levels, rather than maintaining a pre-programmed and fixed basal insulin amount. However, despite this capability, most commercial AID systems continue to rely on traditional bolus-only IOB to display the amount of insulin delivered to the user on their pump interface. ²² As presented in Figure 2 and Table 2, there may be other automated insulin delivery (eg, automated insulin correction for a glucose value above the target range) which may or may not be shown as IOB to the user. This opacity in insulin delivery can lead to confusion, misinterpretation, and in turn, inadequate preparation among individuals with T1D, especially in relation to exercise.

Figure 2.

A visualization of the “two-tap sink” analogy of insulin delivery to highlight differences in IOB and netIOB under different circumstances including: (a) constant basal insulin delivery, (b) constant basal and user-initiated meal bolus, (c) increased basal insulin delivery, (d) decreased basal insulin delivery, (e) basal insulin suspension with user-initiated bolus, and (f) bolus insulin delivery following a basal insulin suspension. Assume the normal basal rate is constant at 1 U/h. (a) Constant basal insulin delivery: In this example, basal insulin delivery is the same as the scheduled rate. Currently, there is a constant slow drip throughout the hour from the basal tap, and the “water level” (i.e., insulin level) in the sink remains constant. IOB and netIOB are both 0 U. (b) Constant basal and user-initiated meal bolus: The individual’s basal rate remains constant and is not impacting the water level in the sink. With a user-initiated meal bolus of 2 U of insulin, there is a surge of water from the bolus tap, and the water level in the sink will rise. As the basal rate is not differing from what is scheduled, both IOB and netIOB become 2 U. (c) Increased basal insulin delivery: Imagine that insulin is increased above scheduled delivery to 2 U/h in response to a high glucose level. The basal tap has now increased the rate of the water drip, more water is being added to the sink, and the water level is rising. Traditional IOB will not account for this increased delivery, and the user will continue to see IOB as 0 U. The increased delivery and water level will be accounted for in netIOB which will show 1 U. By having the ability to see the positive netIOB on the AID system, individuals with T1D may be prompted to initiate proactive management strategies such as carbohydrate feeding, manually initiating a basal reduction, and/or activating the “exercise” pump setting prior to exercise. (d) Decreased basal insulin delivery: Imagine that an individual has undergone a prolonged fast, their glucose levels are trending lower than typical, and as such, the insulin delivery is reduced to 0.5 U/h. This would mean the basal tap drip would slow, and the water in the sink will be below the constant level that is normally maintained. Traditional IOB will remain at 0 U, not showing this decreased delivery and water level, whereas netIOB will show a negative value of −0.5 U. If netIOB is negative for an extended period, this may prompt a user to check ketone levels to avoid euglycemic diabetic ketoacidosis (DKA). This may also influence their future meal dosing strategy or food choices, given the reduction in circulating insulin activity in their body compared with what the body would typically be expecting. (e) Basal insulin suspension without user-initiated bolus: In certain instances, such as when swimming or playing contact sports, an individual with T1D may have to disconnect their insulin pump. Let’s say that an individual is swimming for one hour and suspends and disconnects their pump, essentially completely turning off the basal tap (to 0 U/h) and as a result, lowering the water level in the sink to below the normal constant level. While IOB will still show 0 U, the netIOB will be −1 U. (f) Bolus insulin delivery following a basal insulin suspension: A negative netIOB value following a basal insulin suspension may prompt the user to administer bolus insulin to “cover” or compensate for the reduced insulin. For instance, from example e, if a user sees a netIOB of −1 U, they may be prompted to initiate a 1 U manual bolus, especially if their glucose is rising above target or already well above target. In the analogy of the tap and sink, this would equate to adding a spurt of water from the bolus tap to bring the water level back to baseline, with a netIOB of 0 U. However, the traditional IOB would indicate 1 U of IOB to the user, potentially misleading an individual to believe they have more circulating insulin activity than they truly do.

Remember these examples are simplifications. Given the exponential insulin activity curves of exogenous insulin delivery, as well as different basal insulin needs at different times of day, the netIOB values will likely differ over time.

Table 2.

Numerical Table to Highlight Differences in IOB and netIOB Under Different Circumstances ^* .

Example	Basal rate (U/h)	Bolus (U)	IOB (U)	netIOB (U)
Constant basal insulin delivery	1.0	0	0	0
Constant basal and user-initiated meal bolus	1.0	2	2	2
Increased basal insulin delivery	2.0	0	0	1
Decreased basal insulin delivery	0.5	0	0	−0.5
Basal insulin suspension without user-initiated bolus	0	0	0	−1
Bolus insulin delivery following a basal insulin suspension	0	1	1	0

^*Assume the normal basal rate is constant at 1 U/h.

For instance, any algorithm-based increase in insulin delivery may not be clearly displayed to users in most commercial AID systems, at least not in real-time on the algorithm interface. This has the potential to lead a user to believe that their circulating insulin levels prior to physical activity are less than they truly are, potentially resulting in glucose levels decreasing more than expected or resulting in hypoglycemia. On the contrary, when the AID system decreases insulin delivery below the expected basal rate, users are also not aware of the reduction in circulating insulin levels, which poses its own challenges. For example, if an AID user is physically active for several hours and has not eaten much carbohydrate, the AID system will likely decrease or suspend insulin delivery for a considerable duration. This may influence mealtime insulin requirements later, or worse, increase the risk for euglycemic DKA. The user has no idea how much less insulin has been given during that time frame (a so called “negative” IOB). This can be particularly concerning if an individual with T1D is taking a sodium-glucose transport protein (SGLT2) inhibitor, potentiating euglycemic DKA in the setting of insulin suspension. ²³

What Is netIOB and Why Might It Matter?

The concept of “net” insulin on board (netIOB) is a crucial advancement in the management of T1D. The netIOB displayed to the user accounts for any bolus and adjusted basal (above or below the scheduled basal rate) insulin delivery, thereby providing a more accurate record of insulin delivery. The use of netIOB has been documented in open-source code since 2015, and its incorporation has been effective in both randomized controlled trials and observational studies.^24-28

The goal of estimating bolus IOB, and the more recently coined IOB or Active Insulin, as included in open-loop and AID systems, has been to help inform insulin delivery and glycemic management strategies for individuals with T1D. However, in commercially-available AID systems, there exists a “black box” around the way insulin delivery is calculated and reported, ¹⁹ thereby limiting its usefulness for the user. For over a decade, netIOB has been widely used in open-source AID systems which are used by several thousands of individuals with T1D. ²⁴ Importantly, this method could be adopted by commercial AID systems in the future.

To better visualize the concept of netIOB compared with traditionally reported IOB, one can imagine a sink with two taps (one representing bolus insulin delivery, and the other basal) as an analogy for insulin delivery (Figure 2). The amount of water in the sink at any given time is the netIOB; the drain/spigot is any reduction in the basal insulin delivery relative to the planned or pre-programmed basal rate; a constant drip tap is the basal insulin delivery; and the large tap represents the bolus insulin delivery. The following examples illustrate when IOB and netIOB will likely differ, and perhaps influence T1D management strategies. To maintain simplicity in these examples, we will assume that basal insulin delivery is constant at 1.0 U/h for the entire day.

Regardless of whether an AID system is presenting a positive or negative netIOB value, the transparent reporting of insulin delivery by an AID system can support informed treatment decisions, which is key for people living with diabetes and using these systems. With the recent regulatory clearance of one open-source AID system, ²⁹ which includes the use of netIOB, AID systems reporting insulin delivery using netIOB will likely continue to have a positive impact on diabetes management options.

Active Insulin Versus NetIOB

While IOB is occasionally equated with “active insulin,” we argue that this characterization might be misleading for people with T1D. All insulin administered (whether basal, bolus, or otherwise) should be considered “active” from a physiological standpoint, and thus, it is important to inform users of the cumulative amount of insulin that is physiologically active in their body. When administering insulin subcutaneously, several factors affect its circulating concentration and physiologic action on glucose disposal. These include the true amount of insulin delivered (which may be hindered by kinked/bent cannulas or lipohypertrophy), fluctuations in insulin sensitivity due to factors such as changes in activity levels or illness, and the pharmacokinetics and pharmacodynamics of different synthetic insulin formulations. ⁷ While advancements in technology might eventually allow for subcutaneous insulin sensors capable of measuring true insulin concentrations, this raises additional questions, such as determining what constitutes usual or typical insulin levels for a specific individual. For now, reporting netIOB likely holds the most relevance for patient care and management, and is something that could be displayed to users on an AID system or to users of digital smart pens that can also help users better understand insulin pharmacokinetics. ³⁰ Figure 3 demonstrates how insulin delivery could be visualized to AID system users, including not only netIOB calculations but also a dotted line prediction (as shown in Figure 3b) to reflect the calculated glucose predictions that the AID system has made with all of the information gathered. The combination of netIOB and predictions can provide extended utility to people with T1D and provide additional benefit for commercial AID users in the future.

Figure 3.

Two different AID system user interfaces. Example (a) shows a current AID system which highlights insulin suspension (red shaded bar), margins for glucose ranges (green line at 180 mg/dL, red line at 70 mg/dL), glucose tracing over the past three hours (dotted gray line), current glucose levels (in mg/dL), current glucose trends (black arrow), and bolus only IOB (in U). Example (b) is a proposed user interface which shows the same glucose tracing but highlights additional information to the user including glucose target range 70–180 mg/dl (green shaded region), user-specified glucose target (green dashed line), exercise target (blue dashed line), user-initiated exercise mode (blue shaded region), predicted glucose (grey dotted line), scheduled basal rate (pink dotted line), actual basal insulin delivery (solid pink line), user-initiated boluses (filled purple triangle), system-initiated boluses (open purple triangle), and time of last user bolus and netIOB in U (bottom right).

Note: All illustrations of pump design and display of screens including insulin delivery, netIOB, and so on are illustrative and free to use or adapt for any purpose, including commercial use, without requiring explicit permission.

Conclusions

Since the first use of insulin with a syringe and needle over 100 years ago, diabetes technology has undergone remarkable advancements which have spanned from the introduction of the first portable pump in the late 1970s to the commercial availability of AID systems in September 2016. ¹¹ However, this rapid evolution also warrants a constant need to evaluate the current concepts and terms related to T1D education and management, which includes evaluating the design and functionality of commercial AID systems. Currently, we propose that the commercial AID systems use of bolus-only IOB fails to capture the nuances of insulin delivery adjustments, particularly those related to fluctuations in insulin delivery that are a key feature of AID systems, and occur frequently. The inclusion of netIOB in open-source AID systems has demonstrated to be a promising solution by offering real-time visibility into both increases and decreases in insulin delivery, whether that be through basal or bolus insulin. As technological advancements continue to unfold, the conversation around insulin delivery will likely change. Nevertheless, emphasizing and clearly reporting netIOB in current AID systems provides a practical and near-term tool to improve patient care and help enhance diabetes management.