Understanding insulin and nutrition administration in glycemic control with the right side of the brain

Critically ill patients frequently experience stress-induced hyperglycaemia, leading to increased morbidity and mortality. Glycaemic control (GC) with insulin therapy alone has proven dicult, due to signicant inter- and intra- patient variability in response to insulin therapy. This study reviews the problem and analyses the impact of physiological dynamics and patient variability on outcome glycemia. A graphical model of metabolic dynamics is used to analyse the impact of fundamental glucose ux dynamics on insulin and nutrition administration in the context of maintaining a glycemic goal. It is used to delineate the limits of ability in controlling insulin and/or nutrition administration to achieve safe, effective glycemic control in critical illness in the presence of low insulin sensitivity and high insulin sensitivity variability.

implications for clinical choices on nutrition delivery, and thus glucose uptake, if the appearance/clearance balance is to be maintained -an often overlooked fact in clinical protocols.
This article uses a graphical model demonstrate the problem, delineate the impact of these dynamic interactions on how safe, effective control might be achieved for all patients.
2.0 Methods: 2.1 Fundamental Dynamics: Figure 1 shows the fundamental metabolic physiological dynamics considered [45,46], including primary routes of glucose appearance and delivery, insulin appearance and delivery, and their uptake and use leading to a net blood glucose level. A directly related mathematical model in [46] is validated in clinical glycemic control [47][48][49][50], insulin sensitivity testing [51][52][53][54], and virtual patients [55][56][57]. For this analysis, time-varying and steady state levels of insulin (plasma I(t) leading to interstitial Q(t) à Q SS in steady state), total nutrition (P(t) à P SS ) and glucose level (G(t) à G SS ) are used. Insulin sensitivity, SI, captures patientspeci c ability for insulin mediated glucose uptake.
In particular, at steady state, Q SS approaches a maximum value, Q MAX , typically achieved at infusion rates of 6-8 U/hour in adults [42,43]. Thus, for a given insulin sensitivity level, SI, the maximum rate of insulinmediated glucose removal is limited. In turn, to match glucose appearance and removal, and thus maintain constant blood glucose, the exogenous nutrition rate is also limited to some P MAX . Thus, as insulin sensitivity falls, Q SS à Q MAX and P SS à P MAX , where above P MAX glucose levels will rise, even with more insulin. Insulin saturation thus connects insulin delivery and nutrition delivery, and de ne limits on each if a given glycemic level is to be maintained. Importantly, most clinical protocols ignore nutrition rates in glycemic control [44, [77][78][79]. However, this linkage shows how high nutrition delivery can lead to hyperglycemia in the presence of insulin saturation, when insulin resistance is high (low SI).

A Graphical Model of Glycemic Control and Physiological Dynamics:
Sections 2.1 and 2.2 de ne all necessary terms for a graphical model of how they interact (SI; Q SS and Q MAX , and P SS and P MAX ). This model is shown in Figure 2, including insulin and nutrition inputs with a potential "max" value, determined by patient-speci c insulin saturation and insulin sensitivity, SI, levels. SI is de ned from low (very resistant) to high. In combination, SI, current glucose concentration, G, and effective insulin, Q, yield a variable glucose removal ( Figure 2, lower left) described by the width of the arrow added to a xed uptake by the brain and other non-insulin mediated tissues [45, [80][81][82][83][84][85][86][87][88].
Hence, for a given SI (e.g. Medium in Figure 2), a glucose target (G Target in Figure 2) is met using insulin to manage a given nutrition input. If SI falls, glucose will rise without changing insulin dose. Equally, if SI rises, the insulin required to hold a given glucose level, for the same amount of nutrition, will fall. These behaviours when SI changes are shown in Figure 3 for xed insulin and nutrition, where glucose rises or falls in accord, fully de ning the model.  Figure 4 shows the steady state at given insulin and nutrition doses below their (related) maximum values. In this case, nutrition might be increased, if desired, compared to a goal feed level or any other metric, and insulin could be increased to maintain steady state glycemia at G SS = G TARGET . Figure 5 shows a highly insulin resistant patient with lower SI than Figure 4, with insulin increased to its maximum and Q SS à Q MAX . To maintain a desired glycemia, nutrition is limited to P MAX , which represents the maximum ux of glucose into (and out of) the blood at which G SS = G TARGET can be maintained. Any higher nutrition input (P SS > P MAX ) would raise steady state glucose (G SS ) above the target, as insulin saturation limits insulin-mediated uptake to a maximum level at a given insulin sensitivity (SI), which is shown in Figure 6. Thus, Figures 5-6 show how insulin resistant patients with relatively lower SI will have to receive restricted nutrition inputs to maintain G SS = G TARGET .
In summary, insulin saturation limits effective insulin for insulin-mediated glucose removal. The lower a patient's insulin sensitivity (more resistant), the lower the insulin-mediated glucose removal possible at this upper limit. As such, relatively higher nutrition for this patient will result in higher steady state glycemia above target. Thus, above the Q SS à Q MAX limit ( Figure 5), nutrition inputs must be reduced to avoid hyperglycemia ( Figure 6) beyond a given target level, showing an interaction of physiological dynamics on an often ignored clinical input in restricting glycemia to an intermediate range, which in turn de nes a subset of the most insulin resistant (low SI) patient hours.

The Impact of Metabolic Variability:
ICU patients display signi cant intra-patient metabolic, and thus outcome glycemic, variability [8,37,89]. Figure 7 the outcome effects of variability in insulin sensitivity, . This metabolic variability and/or uncertainty can move glycemia beyond clinically set or desired hyperglycemic (for falling SI) or hypoglycemic (for rising SI) limits as changing SI changes insulin-mediated glucose uptake. The resulting glycemic variability is associated with worsened outcomes [16,27,30,90]. Figure 8 shows how reducing nutrition can reduce and manage the width of glycemic variability, while also reducing insulin requirements. The key outcome is higher insulin and/or nutrition magni es the glycemic variability arising from changes in SI, widening the range of possible blood glucose level outcomes in the presence of SI variability over time. Reducing the resulting glycemic variability thus reduces risk [16,30,31,33] and improves control. This dynamic interaction will affect patients on higher insulin doses more, who are generally those with lower SI and higher glycemic levels.
In summary, glycemic variability is a major hyper-and hypo-glycemia risk factor. Directly managing glycemic variability via reduced nutrition dosing for patients with low SI can reduce risk and improve outcomes. The analysis in Figures 4-6 clearly shows the trade-off between nutrition delivery, insulin saturation, and control to a speci ed blood glucose level. If insulin action was unlimited, where more insulin resulted in more insulin-mediated glucose uptake, nutrition could be independently set and a speci ed blood glucose level achieved as part of any standard titration problem. However, insulin action and effect is saturated [42,43, [58][59][60][61][62][63][64][65][66][67][68][69][70][71][72][73][74][75][76], and increasing insulin doses eventually have less to no effect, except to raise plasma and interstitial circulating insulin levels. Hence, insulin saturation limits insulin-mediated glucose uptake, in turn limiting the total possible carbohydrate delivery to maintain a speci ed blood glucose level ( Figure  5), unless blood glucose is allowed to rise ( Figure 6). Thus, there is a patient speci c, variable upper limit of carbohydrate delivery, which is a function of the patient's insulin sensitivity (SI).
A high SI value indicates larger potential insulin-mediated glucose uptake is possible, and a low value a more resistant patient. Thus, highly insulin resistant patients, typical of ICU patients requiring glycemic control [1][2][3]5], are more likely to need lower nutrition delivery to maintain desirable glucose levels [91].
This outcome matches results showing lower mortality at reduced average nutrition delivery rates compared to current guideline rates [92][93][94], even though only a subset of more insulin resistant patients and/or patient hours might require such reductions from a typical goal nutrition delivery level.
Currently, nutrition delivery is most frequently set to local clinical standards for virtually all glycemic control protocols. Clinical insulin protocols are thus "carbohydrate blind", neither knowing nor accounting for nutritional intake [95,96]. To date, only the model-based STAR and eMPC protocols consider nutrition explicitly [97][98][99][100], and only STAR modulates nutrition delivery in addition to insulin [98][99][100]. STAR and eMPC also directly identify patient-speci c SI using virtual patient models [101,102], and can thus manage the entire trade-off in Figures 4-6 directly.
Clinical Impact: There are two clinical outcomes of Figures 5-6. The rst is the clinical need to include nutrition delivery in glycemic management, whether or not it is controlled. The second is the clinical need to limit and optimise nutrition delivery for more insulin resistant (low SI) patients as a direct part of safely and effectively managing glycemia, where this restriction is increasingly seen as necessary in general [103,104]. As a result, model-based or similar methods to monitor insulin sensitivity (SI) directly could become more necessary to identify who, when, and how much nutrition should be reduced.

Main
Results: The Impact of Metabolic Variability: SI can vary signi cantly between and within critically ill patients [8,37,89]. Figures 6-8 describe the impact of SI variability on glycemic outcome and safety, and demonstrate the potential need to manage nutrition delivery to mitigate hypo-and hyper-glycemic risk. Figure 7 illustrates the risk of SI variability, where critically ill patients have signi cant variability in their hour-hour insulin sensitivity, particularly early in ICU stay [89,105]. Hypoglycemic risk from rising SI ( ) can result in moderate or severe hypoglycemia in up to 10% of hours in the rst 1-3 days of stay, depending on insulin dose [106]. The width of the potential variation is a function of insulin dose and nutrition rate given, where blood glucose level, insulin dose, and nutrition rate together magnify uncertainty in SI. The implication is smaller insulin doses, which can require lower nutrition rates, result in lower glycemic variability ( Figure 8). Hence, more insulin resistant patients (lower SI) (Figures 4-5), with high blood glucose and insulin (and/or nutrition) rates may bene t from short-term reduction of nutrition to reduce both glycemic level and variability, as shown Figure 8.
Clinical Impact: Inter-and intra-patient metabolic variability in insulin sensitivity can signi cantly change the glucose levels resulting from any given insulin dose. The more insulin resistant the patient, the larger the insulin dose required, and thus the wider the resulting glycemic variability range resulting from changes in insulin sensitivity. Reducing nutrition, within clinically accepted ranges, is a means of managing this glycemic variability, and will be necessary for more resistant patients to mitigate avoidable hyper-and hypo-glycemic events resulting from metabolic variability.
Clinically, it is possible to quantify and thus account for this variability, creating an objective means to reduce hypoglycemic and hyperglycemic risk [89,105,[107][108][109][110] 6 illustrate the potential need to reduce or control nutrition delivery in managing glycemia patients with higher insulin resistance (lower SI). Nutrition delivery for the critically ill is an area with signi cant debate concerning the level of nutrition required, how much is practicable to deliver, delivery route, and the impact of macronutrients [92][93][94]104,[111][112][113][114][115][116][117][118][119][120][121]. Recent analyses lean towards a staged approach, increasing nutrition delivery over ICU stay as patient condition improves, thus limiting nutrition delivery per protocol to lower than full or goal feed levels early in stay (e.g. [103,118,120]) when the greatest insulin resistance and variability most typically occur [8,37].
Figures 5-6 illustrate how insulin saturation limits insulin-mediated glucose uptake and thus limits the level of carbohydrate intake a patient can tolerate to avoid excessive hyperglycemia. The more insulin resistant the patient, the lower this value. Hence, at any given hour of stay, any given patient may be able to tolerate more or less carbohydrate nutrition intake than another otherwise similar patient, where a model-based SI value can differentiate such patients.
Currently, only the STAR protocol explicitly modulates insulin and nutrition to manage glycemic level and variability and risk due to Intra-and inter-patient metabolic variability [38,47,122]. Thus, the results in [91], showing mean nutrition during glycemic control met or exceeded leadin nutrition delivery over days 1-3 of ICU stay in a survey of 158 ICUs in 20 countries by Cahill et al [112], were surprising. Equally, a per-patient analysis showed a signi cant spread of maximum tolerated nutrition intake over the 221 patients (21,769 hours) in [91], where these patient-speci c levels rose each day of ICU stay and faster than proposed in [103,118,120]. These results show how directly accounting for insulin sensitivity and insulin saturation, and its variability, can temporarily reduce nutrition delivery, but maximise overall nutrition delivery based on patient-speci c tolerance, resulting in clinically very high levels of nutrition delivery despite controlling it to manage glycemia. More generally, limiting nutrition to improve glycemic control and minimise risk of glycemic variability does not have to limit total nutrition.

Clinical Takeaways:
The overall results show a relatively complex, often ignored, and di cult to measure trade-off. Highly insulin resistant patients are common and create the conditions where reducing nutrition is necessary. More succinctly, high and variable insulin resistance (low and variable SI) requires direct modulation and control of nutrition, in addition to insulin, to minimise the risk of excessive hyperglycemia and hypoglycemia. It can also result in high levels of overall nutrition delivery over the rst 3-5 days of ICU stay, despite temporary reductions. This outcome and approach require model-based or similar methods to monitor insulin sensitivity (SI) directly at the bedside to understand for whom, when, and how much nutrition should be reduced. Overall, controlling nutrition inputs to provide safe, effective and patientspeci c glycemic control is as much about optimising nutrition as it is about glycemic control.

Conclusions:
Hyperglycemic critically ill patients are highly insulin resistant and highly variable. Glycemic control to intermediate or tighter ranges can improve outcomes, but is hard to achieve safely and effectively. This analysis presents a simple pictorial model to illustrate the trade-off between the saturation of insulin action and insulin-mediated glucose uptake, nutrition delivery, and resulting glycemia. It is used to show how highly insulin resistant patients, typical of hyperglycemic critically ill patients, can require reductions in nutrition delivery to maintain a given glycemic level. More speci cally, it illustrates how any given patient has a maximum nutrition delivery rate they can tolerate, which, based on prior studies, is patientspeci c and time varying, indicative of the signi cant metabolic inter-and intra-patient variability common in this cohort. Hence, nutrition control is a necessary aspect of managing the hyperglycemic critically ill patient, both reduction and augmentation.
Clinical takeaways include the need for nutrition modulation in response to patient-speci c condition to mitigate the risk of excessive hyperglycemia and increased risk of hypoglycemia with increasingly high insulin doses. Equally, controlling nutrition inputs to provide safe, effective and patient-speci c glycemic control is as much about optimising nutrition as it is about glycemic control. Emerging model-based methods of patient management and glycemic control offer the opportunity to implement these approaches into regular care. Availability of data and materials: Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
Competing interests: The authors declare that they have no competing interests.      Steady state at insulin and nutrition inputs below their limited maximum values for a typical patient, and the resulting steady state glucose value GSS = GTARGET.

Figure 5
Steady state when insulin saturation limits effective insulin to QMAX and thus the total insulin-mediated removal is limited at Pmax. As a result, nutrition input to maintain the same steady state glucose level (GSS = GTARGET) is limited. Note SI is lower in this gure compared to Figure 4 (using Figure 2) for a highly insulin resistant patient.

Figure 6
When nutrition is increased past the upper limit (PSS > PMAX) for the same patient as in Figure 4, the steady state glucose rises and because insulin is saturated in effect, the width of the glucose removal ux is the same as in Figure 4.

Figure 7
Variability in insulin sensitivity is common and causes signi cant variability in outcome insulin-mediated glucose ux as SI→SI±∆SI, noting the wider arrow left for rising SI and more insulin-mediated glucose ux for a given insulin dose; and smaller arrow right for falling SI and increasing insulin resistance.