A simple insulin-nutrition protocol for tight glycemic control in critical illness: development and protocol comparison

Classification-based deep neural network vs mixture density network models for insulin sensitivity prediction problem

Model-based glycemic control (GC) protocols are used to treat stress-induced hyperglycaemia in intensive care units (ICUs). The STAR (Stochastic-TARgeted) glycemic control protocol - used in clinical practice in several ICUs in New Zealand, Hungary, Belgium, and Malaysia - is a model-based GC protocol using a patient-specific, model-based insulin sensitivity to describe the patient's actual state. Two neural network based methods are defined in this study to predict the patient's insulin sensitivity parameter: a classification deep neural network and a Mixture Density Network based method. Treatment data from three different patient cohorts are used to train the network models. Accuracy of neural network predictions are compared with the current model- based predictions used to guide care. The prediction accuracy was found to be the same or better than the reference. The authors suggest that these methods may be a promising alternative in model-based clinical treatment for patient state prediction. Still, more research is needed to validate these findings, including in-silico simulations and clinical validation trials.

Estimating Enhanced Endogenous Glucose Production in Intensive Care Unit Patients with Severe Insulin Resistance

BACKGROUND

Critically ill ICU patients frequently experience acute insulin resistance and increased endogenous glucose production, manifesting as stress-induced hyperglycemia and hyperinsulinemia. STAR (Stochastic TARgeted) is a glycemic control protocol, which directly manages inter- and intra- patient variability using model-based insulin sensitivity (SI). The model behind STAR assumes a population constant for endogenous glucose production (EGP), which is not otherwise identifiable.

OBJECTIVE

This study analyses the effect of estimating EGP for ICU patients with very low SI (severe insulin resistance) and its impact on identified, model-based insulin sensitivity identification, modeling accuracy, and model-based glycemic clinical control.

METHODS

Using clinical data from 717 STAR patients in 3 independent cohorts (Hungary, New Zealand, and Malaysia), insulin sensitivity, time of insulin resistance, and EGP values are analyzed. A method is presented to estimate EGP in the presence of non-physiologically low SI. Performance is assessed via model accuracy.

RESULTS

Results show 22%-62% of patients experience 1+ episodes of severe insulin resistance, representing 0.87%-9.00% of hours. Episodes primarily occur in the first 24 h, matching clinical expectations. The Malaysian cohort is most affected. In this subset of hours, constant model-based EGP values can bias identified SI and increase blood glucose (BG) fitting error. Using the EGP estimation method presented in these constrained hours significantly reduced BG fitting errors.

CONCLUSIONS

Patients early in ICU stay may have significantly increased EGP. Increasing modeled EGP in model-based glycemic control can improve control accuracy in these hours. The results provide new insight into the frequency and level of significantly increased EGP in critical illness.

Risk and reward: extending stochastic glycaemic control intervals to reduce workload

Background

STAR is a model-based, personalised, risk-based dosing approach for glycaemic control (GC) in critically ill patients. STAR provides safe, effective control to nearly all patients, using 1–3 hourly measurement and intervention intervals. However, the average 11–12 measurements per day required can be a clinical burden in many intensive care units. This study aims to significantly reduce workload by extending STAR 1–3 hourly intervals to 1 to 4-, 5-, and 6-hourly intervals, and evaluate the impact of these longer intervals on GC safety and efficacy, using validated in silico virtual patients and trials methods. A Standard STAR approach was used which allowed more hyperglycaemia over extended intervals, and a STAR Upper Limit Controlled approach limited nutrition to mitigate hyperglycaemia over longer intervention intervals.

Results

Extending STAR from 1–3 hourly to 1–6 hourly provided high safety and efficacy for nearly all patients in both approaches. For STAR Standard, virtual trial results showed lower % blood glucose (BG) in the safe 4.4–8.0 mmol/L target band (from 83 to 80%) as treatment intervals increased. Longer intervals resulted in increased risks of hyper- (15% to 18% BG > 8.0 mmol/L) and hypo- (2.1% to 2.8% of patients with min. BG < 2.2 mmol/L) glycaemia. These results were achieved with slightly reduced insulin (3.2 [2.0 5.0] to 2.5 [1.5 3.0] U/h) and nutrition (100 [85 100] to 90 [75 100] % goal feed) rates, but most importantly, with significantly reduced workload (12 to 8 measurements per day). The STAR Upper Limit Controlled approach mitigated hyperglycaemia and had lower insulin and significantly lower nutrition administration rates.

Conclusions

The modest increased risk of hyper- and hypo-glycaemia, and the reduction in nutrition delivery associated with longer treatment intervals represent a significant risk and reward trade-off in GC. However, STAR still provided highly safe, effective control for nearly all patients regardless of treatment intervals and approach, showing this unique risk-based dosing approach, modulating both insulin and nutrition, to be robust in its design. Clinical pilot trials using STAR with different measurement timeframes should be undertaken to confirm these results clinically.

Simulation Software for Assessment of Nonlinear and Adaptive Multivariable Control Algorithms: Glucose - Insulin Dynamics in Type 1 Diabetes

A simulator for testing automatic control algorithms for nonlinear systems with time-varying parameters, variable time delays, and uncertainties is developed. It is based on simulation of virtual patients with Type 1 diabetes (T1D). Nonlinear models are developed to describe glucose concentration (GC) variations based on user-defined scenarios for meal consumption, insulin administration, and physical activity. They compute GC values and physiological variables, such as heart rate, skin temperature, accelerometer, and energy expenditure, that are indicative of physical activities affecting GC dynamics. This is the first simulator designed for assessment of multivariable controllers that consider supplemental physiological variables in addition to GC measurements to improve glycemic control. Virtual patients are generated from distributions of identified model parameters using clinical data. The simulator will enable testing and evaluation of new control algorithms proposed for automated insulin delivery as well as various control algorithms for nonlinear systems with uncertainties, time-varying parameters and delays.

3D kernel-density stochastic model for more personalized glycaemic control: development and in-silico validation

Background

The challenges of glycaemic control in critically ill patients have been debated for 20 years. While glycaemic control shows benefits inter- and intra-patient metabolic variability results in increased hypoglycaemia and glycaemic variability, both increasing morbidity and mortality. Hence, current recommendations for glycaemic control target higher glycaemic ranges, guided by the fear of harm. Lately, studies have proven the ability to provide safe, effective control for lower, normoglycaemic, ranges, using model-based computerised methods. Such methods usually identify patient-specific physiological parameters to personalize titration of insulin and/or nutrition. The Stochastic-Targeted (STAR) glycaemic control framework uses patient-specific insulin sensitivity and a stochastic model of its future variability to directly account for both inter- and intra-patient variability in a risk-based insulin-dosing approach.

Results

In this study, a more personalized and specific 3D version of the stochastic model used in STAR is compared to the current 2D stochastic model, both built using kernel-density estimation methods. Fivefold cross validation on 681 retrospective patient glycaemic control episodes, totalling over 65,000 h of control, is used to determine whether the 3D model better captures metabolic variability, and the potential gain in glycaemic outcome is assessed using validated virtual trials. Results show that the 3D stochastic model has similar forward predictive power, but provides significantly tighter, more patient-specific, prediction ranges, showing the 2D model over-conservative > 70% of the time. Virtual trial results show that overall glycaemic safety and performance are similar, but the 3D stochastic model reduced median blood glucose levels (6.3 [5.7, 7.0] vs. 6.2 [5.6, 6.9]) with a higher 61% vs. 56% of blood glucose within the 4.4–6.5 mmol/L range.

Conclusions

This improved performance is achieved with higher insulin rates and higher carbohydrate intake, but no loss in safety from hypoglycaemia. Thus, the 3D stochastic model developed better characterises patient-specific future insulin sensitivity dynamics, resulting in improved simulated glycaemic outcomes and a greater level of personalization in control. The results justify inclusion into ongoing clinical use of STAR.

Nutrition delivery, workload and performance in a model-based ICU glycaemic control system

BACKGROUND AND OBJECTIVE

Hyperglycaemia is commonplace in the adult intensive care unit (ICU), and has been associated with increased morbidity and mortality. Effective glycaemic control (GC) can reduce morbidity and mortality, but has proven difficult. STAR is a model-based GC protocol that uniquely maintains normoglycaemia by changing both insulin and nutrition interventions, and has been proven effective in controlling blood glucose (BG) in the ICU. However, most ICU GC protocols only change insulin interventions, making the variable nutrition aspect of STAR less clinically desirable. This paper compares the performance of STAR modulating only insulin, with three simpler alternative nutrition protocols in clinically evaluated virtual trials.

METHODS

Alternative nutrition protocols are fixed nutrition rate (100% caloric goal), CB (Cahill et al. best) stepped nutrition rate (60%, 80% and 100% caloric goal for the first 3 days of GC, and 100% thereafter) and SLQ (STAR lower quartile) stepped nutrition rate (65%, 75% and 85% caloric goal for the first 3 days of GC, and 85% thereafter). Each nutrition protocol is simulated with the STAR insulin protocol on a 221 patient virtual cohort, and GC performance, safety and total intervention workload are assessed.

RESULTS

All alternative nutrition protocols considerably reduced total intervention workload (14.6-19.8%) due to reduced numbers of nutrition changes. However, only the stepped nutrition protocols achieved similar GC performance to the current variable nutrition protocol. Of the two stepped nutrition protocols, the SLQ nutrition protocol also improved GC safety, almost halving the number of severe hypoglycaemic cases (5 vs. 9, P = 0.42).

CONCLUSIONS

Overall, the SLQ nutrition protocol was the best alternative to the current variable nutrition protocol, but either stepped nutrition protocol could be adapted by STAR to reduce workload and make it more clinically acceptable, while maintaining its proven performance and safety.

Improving glycemic control in critically ill patients: personalized care to mimic the endocrine pancreas

There is considerable physiological and clinical evidence of harm and increased risk of death associated with dysglycemia in critical care. However, glycemic control (GC) currently leads to increased hypoglycemia, independently associated with a greater risk of death. Indeed, recent evidence suggests GC is difficult to safely and effectively achieve for all patients. In this review, leading experts in the field discuss this evidence and relevant data in diabetology, including the artificial pancreas, and suggest how safe, effective GC can be achieved in critically ill patients in ways seeking to mimic normal islet cell function. The review is structured around the specific clinical hurdles of: understanding the patient’s metabolic state; designing GC to fit clinical practice, safety, efficacy, and workload; and the need for standardized metrics. These aspects are addressed by reviewing relevant recent advances in science and technology. Finally, we provide a set of concise recommendations to advance the safety, quality, consistency, and clinical uptake of GC in critical care. This review thus presents a roadmap toward better, more personalized metabolic care and improved patient outcomes.

Performance of Stochastic Targeted Blood Glucose Control Protocol by virtual trials in the Malaysian intensive care unit

BACKGROUND AND OBJECTIVE

Blood glucose variability is common in healthcare and it is not related or influenced by diabetes mellitus. To minimise the risk of high blood glucose in critically ill patients, Stochastic Targeted Blood Glucose Control Protocol is used in intensive care unit at hospitals worldwide. Thus, this study focuses on the performance of stochastic modelling protocol in comparison to the current blood glucose management protocols in the Malaysian intensive care unit. Also, this study is to assess the effectiveness of Stochastic Targeted Blood Glucose Control Protocol when it is applied to a cohort of diabetic patients.

METHODS

Retrospective data from 210 patients were obtained from a general hospital in Malaysia from May 2014 until June 2015, where 123 patients were having comorbid diabetes mellitus. The comparison of blood glucose control protocol performance between both protocol simulations was conducted through blood glucose fitted with physiological modelling on top of virtual trial simulations, mean calculation of simulation error and several graphical comparisons using stochastic modelling.

RESULTS

Stochastic Targeted Blood Glucose Control Protocol reduces hyperglycaemia by 16% in diabetic and 9% in nondiabetic cohorts. The protocol helps to control blood glucose level in the targeted range of 4.0-10.0 mmol/L for 71.8% in diabetic and 82.7% in nondiabetic cohorts, besides minimising the treatment hour up to 71 h for 123 diabetic patients and 39 h for 87 nondiabetic patients.

CONCLUSION

It is concluded that Stochastic Targeted Blood Glucose Control Protocol is good in reducing hyperglycaemia as compared to the current blood glucose management protocol in the Malaysian intensive care unit. Hence, the current Malaysian intensive care unit protocols need to be modified to enhance their performance, especially in the integration of insulin and nutrition intervention in decreasing the hyperglycaemia incidences. Improvement in Stochastic Targeted Blood Glucose Control Protocol in terms of u_en model is also a must to adapt with the diabetic cohort.

Effectiveness of an evidence-based protocol for the control of stress-induced hyperglycaemia in critical care

AIM

To assess the effectiveness of the implementation of a protocol for glycaemic control in critical care, in terms of maintenance of a pre-established target of blood glucose level, reduction of hyperglycaemia and prevention of severe hypoglycaemia.

METHOD

Prospective "pre-post" quasi-experimental study carried out in a general critical care unit. Adult patients treated with intravenous insulin were included. We recorded all glycaemic tests performed from November 2014 to August 2015 (pre-intervention) and from November 2015 to August 2016 (post-intervention). The intervention consisted of the implementation of an evidence-based glycaemic control protocol to achieve glycaemic levels in a range of 140-180mg/dl. Main variables analysed were: proportion of glycaemic tests in the target range, proportions of severe hypoglycaemia (under 40mg/dl) and hyperglycaemia over 200mg/dl.

RESULTS

We analysed 7864 glycaemic tests from 125 patients, 66 pre-intervention and 59 post-intervention. Average age was 66.24±13.99 years, 64% of patients were male. The proportion of tests within the target range was higher in the intervention group (38.82 vs. 44.34 p<.001). Only one case of severe hypoglycaemia was identified, which happened in the pre-intervention period. The rate of severe hyperglycaemia was lower in the post-intervention group (19.19 vs. 16.28 p=.001).

CONCLUSIONS

Our experience shows that implementation of evidence-based interventions can improve glycaemic control during critical illness. We found higher glycaemia levels in the target range. The protocol proved useful in the prevention of severe hypoglycaemia. Nurse-led interventions based on clinical data improved health results in our patients.

Next-generation, personalised, model-based critical care medicine: a state-of-the art review of in silico virtual patient models, methods, and cohorts, and how to validation them

Critical care, like many healthcare areas, is under a dual assault from significantly increasing demographic and economic pressures. Intensive care unit (ICU) patients are highly variable in response to treatment, and increasingly aging populations mean ICUs are under increasing demand and their cohorts are increasingly ill. Equally, patient expectations are growing, while the economic ability to deliver care to all is declining. Better, more productive care is thus the big challenge. One means to that end is personalised care designed to manage the significant inter- and intra-patient variability that makes the ICU patient difficult. Thus, moving from current "one size fits all" protocolised care to adaptive, model-based "one method fits all" personalised care could deliver the required step change in the quality, and simultaneously the productivity and cost, of care. Computer models of human physiology are a unique tool to personalise care, as they can couple clinical data with mathematical methods to create subject-specific models and virtual patients to design new, personalised and more optimal protocols, as well as to guide care in real-time. They rely on identifying time varying patient-specific parameters in the model that capture inter- and intra-patient variability, the difference between patients and the evolution of patient condition. Properly validated, virtual patients represent the real patients, and can be used in silico to test different protocols or interventions, or in real-time to guide care. Hence, the underlying models and methods create the foundation for next generation care, as well as a tool for safely and rapidly developing personalised treatment protocols over large virtual cohorts using virtual trials. This review examines the models and methods used to create virtual patients. Specifically, it presents the models types and structures used and the data required. It then covers how to validate the resulting virtual patients and trials, and how these virtual trials can help design and optimise clinical trial. Links between these models and higher order, more complex physiome models are also discussed. In each section, it explores the progress reported up to date, especially on core ICU therapies in glycemic, circulatory and mechanical ventilation management, where high cost and frequency of occurrence provide a significant opportunity for model-based methods to have measurable clinical and economic impact. The outcomes are readily generalised to other areas of medical care.

Generalisability of a Virtual Trials Method for Glycaemic Control in Intensive Care

BACKGROUND

Elevated blood glucose (BG) concentrations (Hyperglycaemia) are a common complication in critically ill patients. Insulin therapy is commonly used to treat hyperglycaemia, but metabolic variability often results in poor BG control and low BG (hypoglycaemia).

OBJECTIVE

This paper presents a model-based virtual trial method for glycaemic control protocol design, and evaluates its generalisability across different populations.

METHODS

Model-based insulin sensitivity (SI) was used to create virtual patients from clinical data from three different ICUs in New Zealand, Hungary, and Belgium. Glycaemic results from simulation of virtual patients under their original protocol (self-simulation) and protocols from other units (cross simulation) were compared.

RESULTS

Differences were found between the three cohorts in median SI and inter-patient variability in SI. However, hour-to-hour intra-patient variability in SI was found to be consistent between cohorts. Self and cross-simulation results were found to have overall similarity and consistency, though results may differ in the first 24-48 h due to different cohort starting BG and underlying SI.

CONCLUSIONS AND SIGNIFICANCE

Virtual patients and the virtual trial method were found to be generalisable across different ICUs. This virtual trial method is useful for in silico protocol design and testing, given an understanding of the underlying assumptions and limitations of this method.

Software-guided insulin dosing: tight glycemic control and decreased glycemic derangements in critically ill patients

OBJECTIVE

To determine whether glycemic derangements are more effectively controlled using software-guided insulin dosing compared with paper-based protocols.

PATIENTS AND METHODS

We prospectively evaluated consecutive critically ill patients treated in a tertiary hospital surgical intensive care unit (ICU) between January 1 and June 30, 2008, and between January 1 and September 30, 2009. Paper-based protocol insulin dosing was evaluated as a baseline during the first period, followed by software-guided insulin dosing in the second period. We compared glycemic metrics related to hyperglycemia, hypoglycemia, and glycemic variability during the 2 periods.

RESULTS

We treated 110 patients by the paper-based protocol and 87 by the software-guided protocol during the before and after periods, respectively. The mean ICU admission blood glucose (BG) level was higher in patients receiving software-guided intensive insulin than for those receiving paper-based intensive insulin (181 vs 156 mg/dL; P=.003, mean of the per-patient mean). Patients treated with software-guided intensive insulin had lower mean BG levels (117 vs 135 mg/dL; P=.0008), sustained greater time in the desired BG target range (95-135 mg/dL; 68% vs 52%; P=.0001), had less frequent hypoglycemia (percentage of time BG level was <70 mg/dL: 0.51% vs 1.44%; P=.04), and showed decreased glycemic variability (BG level per-patient standard deviation from the mean: ±29 vs ±42 mg/dL; P=.01).

CONCLUSION

Surgical ICU patients whose intensive insulin infusions were managed using the software-guided program achieved tighter glycemic control and fewer glycemic derangements than those managed with the paper-based insulin dosing regimen.

Impact of variation in patient response on model-based control of glycaemia in critically ill patients

Critically ill patients commonly experience stress-induced hyperglycaemia, and several studies have shown tight glycaemic control (TGC) can reduce patient mortality. However, tight control is often difficult to achieve due to conflicting drug therapies and evolving patient condition. Thus, a number of studies have failed to achieve consistently safe and effective TGC possibly due to the use of fixed insulin dosing protocols over adaptive patient-specific methods. Model-based targeted glucose control can adapt insulin and dextrose interventions to match identified patient insulin sensitivity. This study explores the impact on glycaemic control of assuming patient response to insulin is constant, as many protocols do, versus time-varying. Validated virtual trial simulations of glucose control were performed on adult and neonatal virtual patient cohorts. Results indicate assumptions of constant insulin sensitivity can lead to six-fold increases in incidence of hypoglycaemia, similar to literature reports and a commonly cited issue preventing increased adoption of TGC in critical care. It is clear that adaptive, patient-specific, approaches are better able to manage inter- and intra-patient variability than typical, fixed protocols.

Pilot study of the SPRINT glycemic control protocol in a Hungarian medical intensive care unit

INTRODUCTION

Stress-induced hyperglycemia increases morbidity and mortality. Tight control can reduce mortality but has proven difficult to achieve. The SPRINT (Specialized Relative Insulin and Nutrition Tables) protocol is the only protocol that reduced both mortality and hypoglycemia by modulating both insulin and nutrition, but it has not been tested in independent hospitals.

METHODS

SPRINT was used for 12 adult intensive care unit patients (949 h) at Kálmán Pándy Hospital (Gyula, Hungary) as a clinical practice assessment. Insulin recommendations (0-6 U/h) were administered via constant infusion rather than bolus delivery. Nutrition was administered per local standard protocol, weaning parenteral to enteral nutrition, but was modulated per SPRINT recommendations. Measurement was every 1 to 2 h, per protocol. Glycemic performance is assessed by percentage of blood glucose (BG) measurements in glycemic bands for the cohort and per patient. Safety from hypoglycemia is assessed by numbers of patients with BG < 2.2 (severe) and %BG < 3.0 and < 4.0 mmol/liter (moderate and light). Clinical effort is assessed by measurements per day. Results are median (interquartile range).

RESULTS

There were 742 measurements over 1088 h of control (16.4 measurements/day), which is similar to clinical SPRINT results (16.2/day). Per-patient hours of control were 65 (50-95) h. Initial per-patient BG was 10.5 (7.9-11.2) mmol/liter. All patients (100%) reached 6.1 mmol/liter. Cohort BG was 6.3 (5.5-7.5) mmol/liter, with 42.2%, 65.1% and 77.6% of BG in the 4.0-6.1, 4.0-7.0, and 4.0-8.0 mmol/liter bands. Per-patient, median percentage time in these bands was 40.2 (26.7-51.5)%, 62.5 (46.0-75.7)%, and 74.7 (61.6.8-87.8)%, respectively. No patients had BG < 2.2 mmol/liter, and the %BG < 4.0 mmol/liter was 1.9%. These results were achieved using 3.0 (3.0-5.0) U/h of insulin with 7.4 (4.4-10.2) g/h of dextrose administration (all sources) for the cohort. Per-patient median insulin administration was 3.0 (3.0-3.0) U/h and 7.1 (3.4-9.6) g/h dextrose. Higher carbohydrate nutrition formulas than were used in SPRINT are offset by slightly higher insulin administration in this study.

CONCLUSIONS

The glycemic performance shows that using the SPRINT protocol to guide insulin infusions and nutrition administration provided very good glycemic control in initial pilot testing, with no severe hypoglycemia. The overall design of the protocol was able to be generalized with good compliance and outcomes across geographically distinct clinical units, patients, and clinical practice.

First pilot trial of the STAR-Liege protocol for tight glycemic control in critically ill patients

Tight glycemic control (TGC) has shown benefits in ICU patients, but been difficult to achieve consistently due to inter- and intra- patient variability that requires more adaptive, patient-specific solutions. STAR (Stochastic TARgeted) is a flexible model-based TGC framework accounting for patient variability with a stochastically derived maximum 5% risk of blood glucose (BG) below 72 mg/dL. This research describes the first clinical pilot trial of the STAR approach and the post-trial analysis of the models and methods that underpin the protocol. The STAR framework works with clinically specified targets and intervention guidelines. The clinically specified glycemic target was 125 mg/dL. Each trial was 24 h with BG measured 1-2 hourly. Two-hourly measurement was used when BG was between 110-135 mg/dL for 3 h. In the STAR approach, each intervention leads to a predicted BG level and outcome range (5-95th percentile) based on a stochastic model of metabolic patient variability. Carbohydrate intake (all sources) was monitored, but not changed from clinical settings except to prevent BG<100 mg/dL when no insulin was given. Insulin infusion rates were limited (6 U/h maximum), with limited increases based on current infusion rate (0.5-2.0 U/h), making this use of the STAR framework an insulin-only TGC approach. Approval was granted by the Ethics Committee of the Medical Faculty of the University of Liege (Liege, Belgium). Nine patient trials were undertaken after obtaining informed consent. There were 205 measurements over all 9 trials. Median [IQR] per-patient results were: BG: 138.5 [130.6-146.0]mg/dL; carbohydrate administered: 2-11 g/h; median insulin:1.3 [0.9-2.4]U/h with a maximum of 6.0 [4.7-6.0]U/h. Median [IQR] time in the desired 110-140 mg/dL band was: 50.0 [31.2-54.2]%. Median model prediction errors ranged: 10-18%, with larger errors due to small meals and other clinical events. The minimum BG was 63 mg/dL and no other measurement was below 72 mg/dL, so only 1 measurement (0.5%) was below the 5% guaranteed minimum risk level. Post-trial analysis showed that patients were more variable than predicted by the stochastic model used for control, resulting in some of the prediction errors seen. Analysis and (validated) virtual trial re-simulating the clinical trial using stochastic models relevant to the patient's particular day of ICU stay were seen to be more accurate in capturing the observed variability. This analysis indicated that equivalent control and safety could be obtained with similar or lower glycemic variability in control using more specific stochastic models. STAR effectively controlled all patients to target. Observed patient variability in response to insulin and thus prediction errors were higher than expected, likely due to the recent insult of cardiac surgery or a major cardiac event, and their immediate recovery. STAR effectively managed this variability with no hypoglycemia. Improved stochastic models will be used to prospectively test these outcomes in further ongoing clinical pilot trials in this and other units.

Pilot study of a model-based approach to blood glucose control in very-low-birthweight neonates

Background

Hyperglycemia often occurs in premature, very low birthweight infants (VLBW) due to immaturity of endogenous regulatory systems and the stress of their condition. Hyperglycemia in neonates has been linked to increased morbidities and mortality and occurs at increasing rates with decreasing birthweight. In this cohort, the emerging use of insulin to manage hyperglycemia has carried a significant risk of hypoglycemia. The efficacy of blood glucose control using a computer metabolic system model to determine insulin infusion rates was assessed in very-low-birth-weight infants.

Methods

Initial short-term 24-hour trials were performed on 8 VLBW infants with hyperglycemia followed by long-term trials of several days performed on 22 infants. Median birthweight was 745 g and 760 g for short-term and long-term trial infants, and median gestational age at birth was 25.6 and 25.4 weeks respectively. Blood glucose control is compared to 21 retrospective patients from the same unit who received insulin infusions determined by sliding scales and clinician intuition. This study was approved by the Upper South A Regional Ethics Committee, New Zealand (ClinicalTrials.gov registration NCT01419873).

Results

Reduction in hyperglycemia towards the target glucose band was achieved safely in all cases during the short-term trials with no hypoglycemic episodes. Lower median blood glucose concentration was achieved during clinical implementation at 6.6 mmol/L (IQR: 5.5 – 8.2 mmol/L, 1,003 measurements), compared to 8.0 mmol/L achieved in similar infants previously (p < 0.01). No significant difference in incidence of hypoglycemia during long-term trials was observed (0.25% vs 0.25%, p = 0.51). Percentage of blood glucose within the 4.0 – 8.0 mmol/L range was increased by 41% compared to the retrospective cohort (68.4% vs 48.4%, p < 0.01).

Conclusions

A computer model that accurately captures the dynamics of neonatal metabolism can provide safe and effective blood glucose control without increasing hypoglycemia.

Trial Registration

ClinicalTrials.gov registration NCT01419873

Variability of insulin sensitivity during the first 4 days of critical illness: implications for tight glycemic control

Background

Effective tight glycemic control (TGC) can improve outcomes in critical care patients, but it is difficult to achieve consistently. Insulin sensitivity defines the metabolic balance between insulin concentration and insulin-mediated glucose disposal. Hence, variability of insulin sensitivity can cause variable glycemia. This study quantifies and compares the daily evolution of insulin sensitivity level and variability for critical care patients receiving TGC.

Methods

This is a retrospective analysis of data from the SPRINT TGC study involving patients admitted to a mixed medical-surgical ICU between August 2005 and May 2007. Only patients who commenced TGC within 12 hours of ICU admission and spent at least 24 hours on the SPRINT protocol were included (N = 164). Model-based insulin sensitivity (SI) was identified each hour. Absolute level and hour-to-hour percent changes in SI were assessed on cohort and per-patient bases. Levels and variability of SI were compared over time on 24-hour and 6-hour timescales for the first 4 days of ICU stay.

Results

Cohort and per-patient median SI levels increased by 34% and 33% (p < 0.001) between days 1 and 2 of ICU stay. Concomitantly, cohort and per-patient SI variability decreased by 32% and 36% (p < 0.001). For 72% of the cohort, median SI on day 2 was higher than on day 1. The day 1–2 results are the only clear, statistically significant trends across both analyses. Analysis of the first 24 hours using 6-hour blocks of SI data showed that most of the improvement in insulin sensitivity level and variability seen between days 1 and 2 occurred during the first 12–18 hours of day 1.

Conclusions

Critically ill patients have significantly lower and more variable insulin sensitivity on day 1 than later in their ICU stay and particularly during the first 12 hours. This rapid improvement is likely due to the decline of counter-regulatory hormones as the acute phase of critical illness progresses. Clinically, these results suggest that while using TGC protocols with patients during their first few days of ICU stay, extra care should be afforded. Increased measurement frequency, higher target glycemic bands, conservative insulin dosing, and modulation of carbohydrate nutrition should be considered to minimize safely the outcome glycemic variability and reduce the risk of hypoglycemia.

Stochastic targeted (STAR) glycemic control: design, safety, and performance

INTRODUCTION

Tight glycemic control (TGC) has shown benefits but has been difficult to achieve consistently. STAR (Stochastic TARgeted) is a flexible, model-based TGC approach that directly accounts for intra- and interpatient variability with a stochastically derived maximum 5% risk of blood glucose (BG) below 72 mg/dl. This research assesses the safety, efficacy, and clinical burden of a STAR TGC controller modulating both insulin and nutrition inputs in virtual and clinical pilot trials.

METHODS

Clinically validated virtual trials using data from 370 patients in the SPRINT (Specialized Relative Insulin and Nutrition Titration) study were used to design the STAR protocol and test its safety, performance, and required clinical effort prior to clinical pilot trials. Insulin and nutrition interventions were given every 1-3 h as chosen by the nurse to allow them to manage workload. Interventions were designed to maximize the overlap of the model-predicted (5-95(th) percentile) range of BG outcomes with the 72-117 mg/dl band and thus provide a maximum 5% risk of BG <72 mg/dl. Interventions were calculated using clinically validated computer models of human metabolism and its variability in critical illness. Carbohydrate intake (all sources) was selected to maximize intake up to 100% of the American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) goal (25 kg/kcal/h). Insulin doses were limited (8 U/h maximum), with limited increases based on current rate (0.5-2.0 U/h). Initial clinical pilot trials involved 3 patients covering ~450 h. Approval was granted by the Upper South A Regional Ethics Committee.

RESULTS

Virtual trials indicate that STAR provides similar glycemic control performance to SPRINT with 2-3 h (maximum) measurement intervals. Time in the 72-126 mg/dl and 72-145 mg/dl bands was equivalent for all controllers, indicating that glycemic outcome differences between protocols were only shifted in this range. Safety from hypoglycemia was improved. Importantly, STAR using 2-3 h (maximum) intervention intervals reduced clinical burden up to 30%, which is clinically very significant. Initial clinical trials showed glycemic performance, safety, and management of inter- and intrapatient variability that matched or exceeded the virtual trial results.

CONCLUSIONS

In virtual trials, STAR TGC provided tight control that maximized the likelihood of BG in a clinically specified glycemic band and reduced hypoglycemia with a maximum 5% (or lower) expected risk of light hypoglycemia (BG <72 mg/dl) via model-based management of intra- and interpatient variability. Clinical workload was self-managed and reduced up to 30% compared with SPRINT. Initial pilot clinical trials matched or exceeded these virtual results.

Pilot proof of concept clinical trials of Stochastic Targeted (STAR) glycemic control

Introduction

Tight glycemic control (TGC) has shown benefits but has been difficult to achieve consistently. STAR (Stochastic TARgeted) is a flexible, model-based TGC approach directly accounting for intra- and inter- patient variability with a stochastically derived maximum 5% risk of blood glucose (BG) < 4.0 mmol/L. This research assesses the safety, efficacy, and clinical burden of a STAR TGC controller modulating both insulin and nutrition inputs in pilot trials.

Methods

Seven patients covering 660 hours. Insulin and nutrition interventions are given 1-3 hourly as chosen by the nurse to allow them to manage workload. Interventions are calculated by using clinically validated computer models of human metabolism and its variability in critical illness to maximize the overlap of the model-predicted (5-95^th percentile) range of BG outcomes with the 4.0-6.5 mmol/L band while ensuring a maximum 5% risk of BG < 4.0 mmol/L. Carbohydrate intake (all sources) was selected to maximize intake up to 100% of SCCM/ACCP goal (25 kg/kcal/h). Maximum insulin doses and dose changes were limited for safety. Measurements were made with glucometers. Results are compared to those for the SPRINT study, which reduced mortality 25-40% for length of stay ≥3 days. Written informed consent was obtained for all patients, and approval was granted by the NZ Upper South A Regional Ethics Committee.

Results

A total of 402 measurements were taken over 660 hours (~14/day), because nurses showed a preference for 2-hourly measurements. Median [interquartile range, (IQR)] cohort BG was 5.9 mmol/L [5.2-6.8]. Overall, 63.2%, 75.9%, and 89.8% of measurements were in the 4.0-6.5, 4.0-7.0, and 4.0-8.0 mmol/L bands. There were no hypoglycemic events (BG < 2.2 mmol/L), and the minimum BG was 3.5 mmol/L with 4.5% < 4.4 mmol/L. Per patient, the median [IQR] hours of TGC was 92 h [29-113] using 53 [19-62] measurements (median, ~13/day). Median [IQR] results: BG, 5.9 mmol/L [5.8-6.3]; carbohydrate nutrition, 6.8 g/h [5.5-8.7] (~70% goal feed median); insulin, 2.5 U/h [0.1-5.1]. All patients achieved BG < 6.1 mmol/L. These results match or exceed SPRINT and clinical workload is reduced more than 20%.

Conclusions

STAR TGC modulating insulin and nutrition inputs provided very tight control with minimal variability by managing intra- and inter- patient variability. Performance and safety exceed that of SPRINT, which reduced mortality and cost in the Christchurch ICU. The use of glucometers did not appear to impact the quality of TGC. Finally, clinical workload was self-managed and reduced 20% compared with SPRINT.

Glycemic control during critical illness

Tight glucose control in the intensive care unit was associated with an improved survival in one pioneering study, but this finding was not confirmed in seven other trials. The reasons underlying this discrepancy are not fully elucidated, indicating that an improved understanding of the physiology and pathophysiology of blood glucose (BG) regulation is needed. Potential reasons for the discrepancies between the studies also include: case-mix, with different optimal BG in different types of patients; hypoglycemia, which is associated with a poor outcome; and the magnitude of glucose variability. Safe and effective glucose control will require progress in the definition of optimal BG, and in the measurement techniques including continuous monitoring insulin algorithms and closed-loop systems.

Glucose control in the intensive care unit: a nutrition support perspective

Hyperglycemia commonly occurs in acutely ill patients who receive nutrition support, even in patients without a history of diabetes. The traditional view that stress hyperglycemia may be a beneficial adaptive response has been replaced by data linking hyperglycemia with increased morbidity and mortality in critically ill populations. Initial randomized studies to control stress hyperglycemia with intensive insulin infusion reported dramatic decreases in infectious complications and decreased mortality. However, recent large multicenter trials have reported that intensive insulin therapy designed to normalize blood glucose resulted in an unacceptable increase in the incidence of hypoglycemia. Review of the methods, protocols, and nutrition provided during these randomized studies is crucial to understanding the different conclusions reached and how these results may be used to influence protocols in intensive care units today. Evidence is reviewed and practical considerations are provided for nutrition support regimens to minimize stress hypoglycemia and assist glucose management.

Evaluating glycemic control algorithms by computer simulations

BACKGROUND

Numerous guidelines and algorithms exist to achieve glycemic control. Their strengths and weaknesses are difficult to assess without head-to-head comparison in time-consuming clinical trials. We hypothesized that computer simulations may be useful.

METHODS

Two open-label randomized clinical trials were replicated using computer simulations. One study compared performance of the enhanced model predictive control (eMPC) algorithm at two intensive care units in the United Kingdom and Belgium. The other study compared three glucose control algorithms-eMPC, Matias (the absolute glucose protocol), and Bath (the relative glucose change protocol)-in a single intensive care unit. Computer simulations utilized a virtual population of 56 critically ill subjects derived from routine data collected at four European surgical and medical intensive care units.

RESULTS

In agreement with the first clinical study, computer simulations reproduced the main finding and discriminated between the two intensive care units in terms of the sampling interval (1.3 h vs. 1.8 h, United Kingdom vs. Belgium; P < 0.01). Other glucose control metrics were comparable between simulations and clinical results. The principal outcome of the second study was also reproduced. The eMPC demonstrated better performance compared with the Matias and Bath algorithms as assessed by the time when plasma glucose was in the target range between 4.4 and 6.1 mmol/L (65% vs. 43% vs. 42% [P < 0.001], eMPC vs. Matias vs. Bath) without increasing the risk of severe hypoglycemia.

CONCLUSIONS

Computer simulations may provide resource-efficient means for preclinical evaluation of algorithms for glycemic control in the critically ill.

Physiological modeling, tight glycemic control, and the ICU clinician: what are models and how can they affect practice?

Critically ill patients are highly variable in their response to care and treatment. This variability and the search for improved outcomes have led to a significant increase in the use of protocolized care to reduce variability in care. However, protocolized care does not address the variability of outcome due to inter- and intra-patient variability, both in physiological state, and the response to disease and treatment. This lack of patient-specificity defines the opportunity for patient-specific approaches to diagnosis, care, and patient management, which are complementary to, and fit within, protocolized approaches.Computational models of human physiology offer the potential, with clinical data, to create patient-specific models that capture a patient's physiological status. Such models can provide new insights into patient condition by turning a series of sometimes confusing clinical data into a clear physiological picture. More directly, they can track patient-specific conditions and thus provide new means of diagnosis and opportunities for optimising therapy.This article presents the concept of model-based therapeutics, the use of computational models in clinical medicine and critical care in specific, as well as its potential clinical advantages, in a format designed for the clinical perspective. The review is presented in terms of a series of questions and answers. These aspects directly address questions concerning what makes a model, how it is made patient-specific, what it can be used for, its limitations and, importantly, what constitutes sufficient validation.To provide a concrete foundation, the concepts are presented broadly, but the details are given in terms of a specific case example. Specifically, tight glycemic control (TGC) is an area where inter- and intra-patient variability can dominate the quality of care control and care received from any given protocol. The overall review clearly shows the concept and significant clinical potential of using computational models in critical care medicine.

Tight glycemic control in critical care--the leading role of insulin sensitivity and patient variability: a review and model-based analysis

Tight glycemic control (TGC) has emerged as a major research focus in critical care due to its potential to simultaneously reduce both mortality and costs. However, repeating initial successful TGC trials that reduced mortality and other outcomes has proven difficult with more failures than successes. Hence, there has been growing debate over the necessity of TGC, its goals, the risk of severe hypoglycemia, and target cohorts. This paper provides a review of TGC via new analyses of data from several clinical trials, including SPRINT, Glucontrol and a recent NICU study. It thus provides both a review of the problem and major background factors driving it, as well as a novel model-based analysis designed to examine these dynamics from a new perspective. Using these clinical results and analysis, the goal is to develop new insights that shed greater light on the leading factors that make TGC difficult and inconsistent, as well as the requirements they thus impose on the design and implementation of TGC protocols. A model-based analysis of insulin sensitivity using data from three different critical care units, comprising over 75,000h of clinical data, is used to analyse variability in metabolic dynamics using a clinically validated model-based insulin sensitivity metric (S(I)). Variation in S(I) provides a new interpretation and explanation for the variable results seen (across cohorts and studies) in applying TGC. In particular, significant intra- and inter-patient variability in insulin resistance (1/S(I)) is seen be a major confounder that makes TGC difficult over diverse cohorts, yielding variable results over many published studies and protocols. Further factors that exacerbate this variability in glycemic outcome are found to include measurement frequency and whether a protocol is blind to carbohydrate administration.

A physiological Intensive Control Insulin-Nutrition-Glucose (ICING) model validated in critically ill patients

Intensive insulin therapy (IIT) and tight glycaemic control (TGC), particularly in intensive care unit (ICU), are the subjects of increasing and controversial debate in recent years. Model-based TGC has shown potential in delivering safe and tight glycaemic management, all the while limiting hypoglycaemia. A comprehensive, more physiologically relevant Intensive Control Insulin-Nutrition-Glucose (ICING) model is presented and validated using data from critically ill patients. Two existing glucose-insulin models are reviewed and formed the basis for the ICING model. Model limitations are discussed with respect to relevant physiology, pharmacodynamics and TGC practicality. Model identifiability issues are carefully considered for clinical settings. This article also contains significant reference to relevant physiology and clinical literature, as well as some references to the modeling efforts in this field. Identification of critical constant population parameters was performed in two stages, thus addressing model identifiability issues. Model predictive performance is the primary factor for optimizing population parameter values. The use of population values are necessary due to the limited clinical data available at the bedside in the clinical control scenario. Insulin sensitivity, S(I), the only dynamic, time-varying parameter, is identified hourly for each individual. All population parameters are justified physiologically and with respect to values reported in the clinical literature. A parameter sensitivity study confirms the validity of limiting time-varying parameters to S(I) only, as well as the choices for the population parameters. The ICING model achieves median fitting error of <1% over data from 173 patients (N=42,941 h in total) who received insulin while in the ICU and stayed for ≥ 72 h. Most importantly, the median per-patient 1-h ahead prediction error is a very low 2.80% [IQR 1.18, 6.41%]. It is significant that the 75th percentile prediction error is within the lower bound of typical glucometer measurement errors of 7-12%. These results confirm that the ICING model is suitable for developing model-based insulin therapies, and capable of delivering real-time model-based TGC with a very tight prediction error range. Finally, the detailed examination and discussion of issues surrounding model-based TGC and existing glucose-insulin models render this article a mini-review of the state of model-based TGC in critical care.

Development of a model-based clinical sepsis biomarker for critically ill patients

Sepsis occurs frequently in the intensive care unit (ICU) and is a leading cause of admission, mortality, and cost. Treatment guidelines recommend early intervention, however positive blood culture results may take up to 48 h. Insulin sensitivity (S(I)) is known to decrease with worsening condition and could thus be used to aid diagnosis. Some glycemic control protocols are able to accurately identify insulin sensitivity in real-time. Hourly model-based insulin sensitivity S(I) values were calculated from glycemic control data of 36 patients with sepsis. The hourly S(I) is compared to the hourly sepsis score (ss) for these patients (ss=0-4 for increasing severity). A multivariate clinical biomarker was also developed to maximize the discrimination between different ss groups. Receiver operator characteristic (ROC) curves for severe sepsis (ss ≥ 2) are created for both S(I) and the multivariate clinical biomarker. Insulin sensitivity as a sepsis biomarker for diagnosis of severe sepsis achieves a 50% sensitivity, 76% specificity, 4.8% positive predictive value (PPV), and 98.3% negative predictive value (NPV) at an S(I) cut-off value of 0.00013 L/mU/min. Multivariate clinical biomarker combining S(I), temperature, heart rate, respiratory rate, blood pressure, and their respective hourly rates of change achieves 73% sensitivity, 80% specificity, 8.4% PPV, and 99.2% NPV. Thus, the multivariate clinical biomarker provides an effective real-time negative predictive diagnostic for severe sepsis. Examination of both inter- and intra-patient statistical distribution of this biomarker and sepsis score shows potential avenues to improve the positive predictive value.

Development of blood glucose control for extremely premature infants

Extremely premature neonates often experience hyperglycaemia, which has been linked to increased mortality and worsened outcomes. Insulin therapy can assist in controlling blood glucose levels and promoting needed growth. This study presents the development of a model-based stochastic targeted controller designed to adapt insulin infusion rates to match the unique and changing metabolic state and control parameters of the neonate. Long-term usage of targeted BG control requires successfully forecasting variations in neonatal metabolic state, accounting for differences in clinical practices between units, and demonstrating robustness to errors that can occur in everyday clinical usage. Simulation studies were used to evaluate controller ability to target several common BG ranges and evaluate controller sensitivity to missed BG measurements and delays in control interventions on a virtual patient cohort of 25 infants developed from retrospective data. Initial clinical pilot trials indicated model performance matched expected performance from simulations. Stochastic targeted glucose control developed using validated patient-specific virtual trials can yield effective protocols for this cohort. Long-term trials show fundamental success, however clinical interface design appears as a critical factor to ensuring good compliance and thus good control.

Impact of glucocorticoids on insulin resistance in the critically ill

Glucocorticoids (GCs) have been shown to reduce insulin sensitivity in healthy individuals. Widely used in critical care to treat a variety of inflammatory and allergic disorders, they may inadvertently exacerbate stress-hyperglycaemia. This research uses model-based methods to quantify the reduction in insulin sensitivity from GCs in critically ill patients, and thus their impact on glycaemic control. A model-based measure of insulin sensitivity (S(I)) was used to quantify changes between two matched cohorts of 40 intensive care unit (ICU) patients. Patients in one cohort received GC treatment, while patients in the control cohort did not. All patients were admitted to the Christchurch hospital ICU between 2005 and 2007 and spent at least 24h on the SPRINT glycaemic control protocol. A 31% reduction in whole-cohort median insulin sensitivity was seen between the control cohort and patients receiving glucocorticoids with a median dose equivalent to 200mg/d of hydrocortisone per patient. Comparing percentile patients as a surrogate for matched patients, reductions in median insulin sensitivity of 20%, 25%, and 21% were observed for the 25th-, 50th- and 75th-percentile patients, respectively. These cohort and percentile patient reductions are less than or equivalent to the 30-62% reductions reported in healthy subjects especially when considering the fact that the GC doses in this study are 1.3-4.0 times larger than those in studies of healthy subjects. This reduced suppression of insulin sensitivity in critically ill patients could be a result of saturation due to already increased levels of catecholamines and cortisol common in critically illness. Virtual trial simulation showed that reductions in insulin sensitivity of 20-30% associated with glucocorticoid treatment in the ICU have limited impact on glycaemic control levels within the context of the SPRINT protocol.

Intravascular glucose/lactate sensors prepared with nitric oxide releasing poly(lactide-co-glycolide)-based coatings for enhanced biocompatibility

Intravenous amperometric needle-type enzymatic glucose/lactate sensors intended for continuous monitoring are prepared with a novel nitric oxide (NO) releasing layer to improve device hemocompatibility. To create an underlying NO release coating, the sensors with immobilized enzymes (either glucose oxidase or lactate oxidase) are prepared with a thin layer of poly(lactide-co-glycolide) (PLGA) loaded with lipophilic diazeniumdiolate species that slowly release NO via a proton driven reaction. An outer thin layer (ca. 30 μm) of PurSil (polyurethane/dimethylsiloxane copolymer) limits the flux of glucose and lactate to the inner layer of enzyme, to provide the desired linear amperometric response. A 30 μm coating of PLGA containing 33 wt% of the appropriate NO donor (N-diazeniumdiolated dibutylhexanediamine, DBHD/N₂O₂) can release NO at a physiologically relevant rate > 1 × 10⁻¹⁰mol min⁻¹ cm⁻² for at least 7 days without influencing the analytical performance of the glucose/lactate sensors. In vitro, the sensors exhibit relatively stable amperometric response over a one-week period with high selectivity over interferences (e.g., ascorbic acid) required for blood monitoring applications. Glucose sensors implanted in the veins of rabbits for 8h exhibit significantly enhanced hemocompatibility for the NO release sensors vs. corresponding controls (without NO release in same animals), with greatly reduced thrombus formation on their surfaces. Further, the analytical performance of the NO release glucose sensors are superior to controls placed in the veins of the same animals, with a greater accuracy in measuring blood glucose levels as evaluated using a Clarke error grid type analysis.

Glucose and insulin administration while maintaining normoglycemia during cardiac surgery using a computer-assisted algorithm

BACKGROUND

applying the principles of the hyperinsulinemic-normoglycemic clamp technique we have introduced glucose and insulin administration while maintaining normoglycemia (GIN therapy) to surgical patients. The objective of this study was to evaluate a novel computer software (GIN Computer Software [GINCS]) program using an algorithm based on the original clamp equation and modified for its use during cardiac surgery.

METHODS

thirty-six patients without diabetes undergoing elective cardiac surgery were randomly assigned to manually controlled or computer-guided GIN therapy. In both groups insulin was administered at 5 mU/kg/min during surgery. Simultaneously, 20% dextrose was infused at a rate adjusted to maintain blood glucose (BG) between 4.0 and 6.0 mmol/L. The adjustments were made either following an algorithm based on our previous GIN experience or suggestions made by the software program. The primary outcome was the achievement of target glycemia.

RESULTS

normoglycemia was achieved in both groups as reflected by mean BG concentrations of 5.0 ± 0.5 mmol/L and 5.1 ± 0.2 mmol/L. Mean sampling intervals were longer in the GINCS group than in the manual group (21.5  ± 1.9 vs. 14.2 ± 2.2 min, P < 0.001). The GINCS therapy was associated with a greater percentage of BG measurements within target (manual group, before cardiopulmonary bypass [CPB] 79.7%, during CPB 68.1%, and after CPB 69.1%; GINCS group, before CPB 94.1%, during CPB 92.4%, and after CPB 97.7%; P < 0.001). No hypoglycemia was observed.

CONCLUSIONS

the use of a computer-guided GIN protocol in patients without diabetes undergoing open heart surgery provided excellent and safe glycemic control.

Validation of a model-based virtual trials method for tight glycemic control in intensive care

BACKGROUND

In-silico virtual patients and trials offer significant advantages in cost, time and safety for designing effective tight glycemic control (TGC) protocols. However, no such method has fully validated the independence of virtual patients (or resulting clinical trial predictions) from the data used to create them. This study uses matched cohorts from a TGC clinical trial to validate virtual patients and in-silico virtual trial models and methods.

METHODS

Data from a 211 patient subset of the Glucontrol trial in Liege, Belgium. Glucontrol-A (N = 142) targeted 4.4-6.1 mmol/L and Glucontrol-B (N = 69) targeted 7.8-10.0 mmol/L. Cohorts were matched by APACHE II score, initial BG, age, weight, BMI and sex (p > 0.25). Virtual patients are created by fitting a clinically validated model to clinical data, yielding time varying insulin sensitivity profiles (SI(t)) that drives in-silico patients.Model fit and intra-patient (forward) prediction errors are used to validate individual in-silico virtual patients. Self-validation (tests A protocol on Group-A virtual patients; and B protocol on B virtual patients) and cross-validation (tests A protocol on Group-B virtual patients; and B protocol on A virtual patients) are used in comparison to clinical data to assess ability to predict clinical trial results.

RESULTS

Model fit errors were small (<0.25%) for all patients, indicating model fitness. Median forward prediction errors were: 4.3, 2.8 and 3.5% for Group-A, Group-B and Overall (A+B), indicating individual virtual patients were accurate representations of real patients. SI and its variability were similar between cohorts indicating they were metabolically similar.Self and cross validation results were within 1-10% of the clinical data for both Group-A and Group-B. Self-validation indicated clinically insignificant errors due to model and/or clinical compliance. Cross-validation clearly showed that virtual patients enabled by identified patient-specific SI(t) profiles can accurately predict the performance of independent and different TGC protocols.

CONCLUSIONS

This study fully validates these virtual patients and in silico virtual trial methods, and clearly shows they can accurately simulate, in advance, the clinical results of a TGC protocol, enabling rapid in silico protocol design and optimization. These outcomes provide the first rigorous validation of a virtual in-silico patient and virtual trials methodology.

Organ failure and tight glycemic control in the SPRINT study

Introduction

Intensive care unit mortality is strongly associated with organ failure rate and severity. The sequential organ failure assessment (SOFA) score is used to evaluate the impact of a successful tight glycemic control (TGC) intervention (SPRINT) on organ failure, morbidity, and thus mortality.

Methods

A retrospective analysis of 371 patients (3,356 days) on SPRINT (August 2005 - April 2007) and 413 retrospective patients (3,211 days) from two years prior, matched by Acute Physiology and Chronic Health Evaluation (APACHE) III. SOFA is calculated daily for each patient. The effect of the SPRINT TGC intervention is assessed by comparing the percentage of patients with SOFA ≤5 each day and its trends over time and cohort/group. Organ-failure free days (all SOFA components ≤2) and number of organ failures (SOFA components >2) are also compared. Cumulative time in 4.0 to 7.0 mmol/L band (cTIB) was evaluated daily to link tightness and consistency of TGC (cTIB ≥0.5) to SOFA ≤5 using conditional and joint probabilities.

Results

Admission and maximum SOFA scores were similar (P = 0.20; P = 0.76), with similar time to maximum (median: one day; IQR: [1,3] days; P = 0.99). Median length of stay was similar (4.1 days SPRINT and 3.8 days Pre-SPRINT; P = 0.94). The percentage of patients with SOFA ≤5 is different over the first 14 days (P = 0.016), rising to approximately 75% for Pre-SPRINT and approximately 85% for SPRINT, with clear separation after two days. Organ-failure-free days were different (SPRINT = 41.6%; Pre-SPRINT = 36.5%; P < 0.0001) as were the percent of total possible organ failures (SPRINT = 16.0%; Pre-SPRINT = 19.0%; P < 0.0001). By Day 3 over 90% of SPRINT patients had cTIB ≥0.5 (37% Pre-SPRINT) reaching 100% by Day 7 (50% Pre-SPRINT). Conditional and joint probabilities indicate tighter, more consistent TGC under SPRINT (cTIB ≥0.5) increased the likelihood SOFA ≤5.

Conclusions

SPRINT TGC resolved organ failure faster, and for more patients, from similar admission and maximum SOFA scores, than conventional control. These reductions mirror the reduced mortality with SPRINT. The cTIB ≥0.5 metric provides a first benchmark linking TGC quality to organ failure. These results support other physiological and clinical results indicating the role tight, consistent TGC can play in reducing organ failure, morbidity and mortality, and should be validated on data from randomised trials.

A fast generalizable solution method for glucose control algorithms

In critical care tight control of blood glucose levels has been shown to lead to better clinical outcomes. The need to develop new protocols for tight glucose control, as well as the opportunity to optimize a variety of other drug therapies, has led to resurgence in model-based medical decision support in this area. One still valid hindrance to developing new model-based protocols using so-called virtual patients, retrospective clinical data, and Monte Carlo methods is the large amount of computational time and resources needed. This paper develops fast analytical-based methods for insulin-glucose system model that are generalizable to other similar systems. Exploiting the structure and partial solutions in a subset of the model is the key in finding accurate fast solutions to the full model. This approach successfully reduced computing time by factors of 5600-144000 depending on the numerical error management method, for large (50-164 patients) virtual trials and Monte Carlo analysis. It thus allows new model-based or model-derived protocols to be rapidly developed via extensive simulation. The new method is rigorously compared to existing standard numerical solutions and is found to be highly accurate to within 0.2%.

Implementing glucose control in intensive care: a multicenter trial using statistical process control

Background

Glucose control (GC) with insulin decreases morbidity and mortality of critically ill patients. In this study we investigated GC performance over time during implementation of GC strategies within three intensive care units (ICUs) and in routine clinical practice.

Methods

All adult critically ill patients who stayed for >24 h between 1999 and 2007 were included. Effects of implementing local GC guidelines and guideline revisions on effectiveness/efficiency-related indicators, safety-related indicators, and protocol-related indicators were measured.

Results

Data of 17,111 patient admissions were evaluated, with 714,141 available blood glucose levels (BGL) measurements. Mean BGL, time to reach target, hyperglycemia index, sampling frequency, percentage of hyperglycemia events, and in-range measurements statistically changed after introducing GC in all ICUs. The introduction of simple rules on GC had the largest effect. Subsequent changes in the protocol had a smaller effect than the introduction of the protocol itself. As soon as the protocol was introduced, in all ICUs the percentage of hypoglycemia events increased. Various revisions were implemented to reduce hypoglycemia events, but levels never returned to those from pre-implementation. More intensive implementation strategies including the use of a decision support system resulted in better control of the process.

Conclusion

There are various strategies to achieve GC in routine clinical practice but with variable success. All of them were associated with an increase in hypoglycemia events, but GC was never stopped. Instead, these events have been accepted and managed. Statistical process control is a useful tool for monitoring phenomena over time and captures within-institution changes.

Electronic supplementary material

The online version of this article (doi:10.1007/s00134-010-1924-3) contains supplementary material, which is available to authorized users.

Continuous glucose monitors and the burden of tight glycemic control in critical care: can they cure the time cost?

BACKGROUND

Tight glycemic control (TGC) in critical care has shown distinct benefits but has also proven to be difficult to obtain. The risk of severe hypoglycemia (<40 mg/dl) raises significant concerns for safety. Added clinical burden has also been an issue. Continuous glucose monitors (CGMs) offer frequent automated measurement and thus the possibility of using them for early detection and intervention of hypoglycemic events. Additionally, regular measurement by CGM may also be able to reduce clinical burden.

AIM

An in silico study investigates the potential of CGM devices to reduce clinical effort in a published TGC protocol.

METHODS

This study uses retrospective clinical data from the Specialized Relative Insulin Nutrition Titration (SPRINT) TGC study covering 20 patients from a benchmark cohort. Clinically validated metabolic system models are used to generate a blood glucose (BG) profile for each patient, resulting in 33 continuous, separate BG episodes (6881 patient hours). The in silico analysis is performed with three different stochastic noise models: two Gaussian and one first-order autoregressive. The noisy, virtual CGM BG values are filtered and used to drive the SPRINT TGC protocol. A simple threshold alarm is used to trigger glucose interventions to avert potential hypoglycemia. The Monte Carlo method was used to get robust results from the stochastic noise models.

RESULTS

Using SPRINT with simulated CGM noise, the BG time in an 80-110 mg/dl band was reduced no more than 4.4% to 45.2% compared to glucometer sensors. Antihypoglycemic interventions had negligible effect on time in band but eliminated all recorded hypoglycemic episodes in these simulations. Assuming 4-6 calibration measurements per day, the nonautomated clinical measurements are reduced from an average of 16 per day to as low as 4. At 2.5 min per glucometer measurement, a daily saving of approximately 25-30 min per patient could potentially be achieved.

CONCLUSIONS

This paper has analyzed in silico the use of CGM sensors to provide BG input data to the SPRINT TGC protocol. A very simple algorithm was used for early hypoglycemic detection and prevention and tested with four different-sized intravenous glucose boluses. Although a small decrease in time in band (still clinically acceptable) was experienced with the addition of CGM noise, the number of hypoglycemic events was reduced. The reduction to time in band depends on the specific CGM sensor error characteristics and is thus a trade-off for reduced nursing workload. These results justify a pilot clinical trial to verify this study.

A simulation model of insulin saturation and glucose balance for glycemic control in ICU patients

Consistent tight blood sugar control in critically ill patients has proven elusive. Properly accounting for the saturation of insulin action and reducing the need for frequent measurements are important aspects in intensive insulin therapy. This paper presents a composite metabolic model, 'Glucosafe', that integrates models and parameters from normal physiology and accounts for the reduced rate of glucose gut absorption and saturation of insulin action in patients with reduced insulin sensitivity. Particularly, two different sites of reduced insulin sensitivity, before and after the non-linearity of insulin action, are explored with this model. These approaches are assessed based on the model's accuracy in retrospectively predicting blood glucose measurements of 10 randomly chosen, hyperglycemic intensive care patients. For each patient, median absolute percent error is <25% for prediction times < or = 270min and modelling reduced insulin sensitivity after the non-linearity, compared to <29% for modelling reduced insulin sensitivity before the non-linearity. Scaling the insulin effect (after the non-linearity) is a suitable assumption in this model structure. These results are preliminary and subject to further and more extensive validation of the model's capability to predict the longer term (>2h) blood glucose excursion in critically ill patients.

What makes tight glycemic control tight? The impact of variability and nutrition in two clinical studies

INTRODUCTION

Tight glycemic control (TGC) remains controversial while successful, consistent, and effective protocols remain elusive. This research analyzes data from two TGC trials for root causes of the differences achieved in control and thus potentially in glycemic and other outcomes. The goal is to uncover aspects of successful TGC and delineate the impact of differences in cohorts.

METHODS

A retrospective analysis was conducted using records from a 211-patient subset of the GluControl trial taken in Liege, Belgium, and 393 patients from Specialized Relative Insulin Nutrition Titration (SPRINT) in New Zealand. Specialized Relative Insulin Nutrition Titration targeted 4.0-6.0 mmol/liter, similar to the GluControl A (N = 142) target of 4.4-6.1 mmol/liter. The GluControl B (N = 69) target was 7.8-10.0 mmol/liter. Cohorts were matched by Acute Physiology and Chronic Health Evaluation II score and percentage males (p > .35); however, the GluControl cohort was slightly older (p = .011). Overall cohort and per-patient comparisons (median, interquartile range) are shown for (a) glycemic levels achieved, (b) nutrition from carbohydrate (all sources), and (c) insulin dosing for this analysis. Intra- and interpatient variability were examined using clinically validated model-based insulin sensitivity metric and its hour-to-hour variation.

RESULTS

Cohort blood glucose were as follows: SPRINT, 5.7 (5.0-6.6) mmol/liter; GluControl A, 6.3 (5.3-7.6) mmol/liter; and GluControl B, 8.2 (6.9-9.4) mmol/liter. Insulin dosing was 3.0 (1.0-3.0), 1.5 (0.5-3), and 0.7 (0.0-1.7) U/h, respectively. Nutrition from carbohydrate (all sources) was 435.5 (259.2-539.1), 311.0 (0.0-933.1), and 622.1 (103.7-1036.8) kcal/day, respectively. Median per-patient results for blood glucose were 5.8 (5.3-6.4), 6.4 (5.9-6.9), and 8.3 (7.6-8.8) mmol/liter. Insulin doses were 3.0 (2.0-3.0), 1.5 (0.8-2.0), and 0.5 (0.0-1.0) U/h. Carbohydrate administration was 383.6 (207.4-497.7), 103.7 (0.0-829.4), and 207.4 (0.0-725.8) kcal/day. Overall, SPRINT gave approximately 2x more insulin with a 3-4x narrower, but generally non-zero, range of nutritional input to achieve equally TGC with less hypoglycemia. Specialized Relative Insulin Nutrition Titration had much less hypoglycemia (<2.2 mmol/liter), with 2% of patients, compared to GluControl A (7.7%) and GluControl B (2.9%), indicating much lower variability, with similar results for glucose levels <3.0 mmol/liter. Specialized Relative Insulin Nutrition Titration also had less hyperglycemia (>8.0 mmol/liter) than groups A and B. GluControl patients (A+B) had a approximately 2x wider range of insulin sensitivity than SPRINT. Hour-to-hour variation was similar. Hence GluControl had greater interpatient variability but similar intrapatient variability.

CONCLUSION

Protocols that dose insulin blind to carbohydrate administration can suffer greater outcome glycemic variability, even if average cohort glycemic targets are met. While the cohorts varied significantly in model-assessed insulin resistance, their variability was similar. Such significant intra- and interpatient variability is a further significant cause and marker of glycemic variability in TGC. The results strongly recommended that TGC protocols be explicitly designed to account for significant intra- and interpatient variability in insulin resistance, as well as specifying or having knowledge of carbohydrate administration to minimize variability in glycemic outcomes across diverse cohorts and/or centers.

Hypoglycemia detection in critical care using continuous glucose monitors: an in silico proof of concept analysis

BACKGROUND

Tight glycemic control (TGC) in critical care has shown distinct benefits but has also been proven difficult to obtain. The risk of severe hypoglycemia (<40 mg/dl) has been increased significantly in several, but not all, studies, raising significant concerns for safety. Continuous glucose monitors (CGMs) offer frequent measurement and thus the possibility of using them for early detection alarms to prevent hypoglycemia.

METHODS

This study used retrospective clinical data from the Specialized Relative Insulin Nutrition Titration TGC study covering seven patients who experienced severe hypoglycemic events. Clinically validated metabolic system models were used to recreate a continuous blood glucose profile. In silico analysis was enabled by using a conservative single Gaussian noise model based on reported CGM clinical data from a critical care study [mean absolute percent error (MAPE) 17.4%]. A novel median filter was implemented and further smoothed with a least mean squares-fitted polynomial to reduce sensor noise. Two alarm approaches were compared. An integral-based method is presented that examined the area between a preset threshold and filtered simulated CGM data. An alarm was raised when this value became too low. A simple glycemic threshold method was also used for comparison. To account for random noise skewing the results, each patient record was Monte Carlo simulated 100 times with a different random noise profile for a total of 700 runs. Different alarm thresholds were analyzed parametrically. Results are reported in terms of detection time before the clinically measured event and any false alarms. These retrospective clinical data were used with approval from the New Zealand South Island Regional Ethics Committee.

RESULTS

The median filter reduced MAPE from 17.4% [standard deviation (SD) 13%] to 9.3% (SD 7%) over the cohort. For the integral-based alarm, median per-patient detection times ranged, t, from -35 minutes (before event) to -170 minutes, with zero to two false alarms per patient over the cohort and different alarm parameters. For a simple glycemic threshold alarm (three consecutive values below threshold), median per-patient alarm times were -10 to -75 minutes and false alarms were zero to seven; however, in one case, five of seven subjects never alarmed at all, despite the hypoglycemic event.

CONCLUSIONS

A retrospective study used clinical hypoglycemic events from a TGC study to develop and analyze an integral-based hypoglycemia alarm for use in critical care TGC studies. The integral-based approach was accurate, provided significant lead time before a hypoglycemic event, alarmed at higher glycemic levels, was robust to sensor noise, and had minimal false alarms. The approach is readily generalizable to similar scenarios, and results would justify a pilot clinical trial to verify this study.

Simulation environment to evaluate closed-loop insulin delivery systems in type 1 diabetes

BACKGROUND

Closed-loop insulin delivery systems linking subcutaneous insulin infusion to real-time continuous glucose monitoring need to be evaluated in humans, but progress can be accelerated with the use of in silico testing. We present a simulation environment designed to support the development and testing of closed-loop insulin delivery systems in type 1 diabetes mellitus (T1DM).

METHODS

The principal components of the simulation environment include a mathematical model of glucose regulation representing a virtual population with T1DM, the glucose measurement model, and the insulin delivery model. The simulation environment is highly flexible. The user can specify an experimental protocol, define a population of virtual subjects, choose glucose measurement and insulin delivery models, and specify outcome measures. The environment provides graphical as well as numerical outputs to enable a comprehensive analysis of in silico study results. The simulation environment is validated by comparing its predictions against a clinical study evaluating overnight closed-loop insulin delivery in young people with T1DM using a model predictive controller.

RESULTS

The simulation model of glucose regulation is described, and population values of 18 synthetic subjects are provided. The validation study demonstrated that the simulation environment was able to reproduce the population results of the clinical study conducted in young people with T1DM.

CONCLUSIONS

Closed-loop trials in humans should be preceded and concurrently guided by highly efficient and resource-saving computer-based simulations. We demonstrate validity of population-based predictions obtained with our simulation environment.

Blood glucose controller for neonatal intensive care: virtual trials development and first clinical trials

BACKGROUND

Premature neonates often experience hyperglycemia, which has been linked to worsened outcomes. Insulin therapy can assist in controlling blood glucose (BG) levels. However, a reliable, robust control protocol is required to avoid hypoglycemia and to ensure that clinically important nutrition goals are met.

METHODS

This study presents an adaptive, model-based predictive controller designed to incorporate the unique metabolic state of the neonate. Controller performance was tested and refined in virtual trials on a 25-patient retrospective cohort. The effects of measurement frequency and BG sensor error were evaluated. A stochastic model of insulin sensitivity was used in control to provide a guaranteed maximum 4% risk of BG < 72 mg/dl to protect against hypoglycemia as well as account for patient variability over 1-3 h intervals when determining the intervention. The resulting controller is demonstrated in two 24 h clinical neonatal pilot trials at Christchurch Women's Hospital.

RESULTS

Time in the 72-126 mg/dl BG band was increased by 103-161% compared to retrospective clinical control for virtual trials of the controller, with fewer hypoglycemic measurements. Controllers were robust to BG sensor errors. The model-based controller maintained glycemia to a tight target control range and accounted for interpatient variability in patient glycemic response despite using more insulin than the retrospective case, illustrating a further measure of controller robustness. Pilot clinical trials demonstrated initial safety and efficacy of the control method.

CONCLUSIONS

A controller was developed that made optimum use of the very limited available BG measurements in the neonatal intensive care unit and provided robustness against BG sensor error and longer BG measurement intervals. It used more insulin than typical sliding scale approaches or retrospective hospital control. The potential advantages of a model-based approach demonstrated in simulation were applied to initial clinical trials.