Reducing the impact of insulin sensitivity variability on glycaemic outcomes using separate stochastic models within the STAR glycaemic protocol

Abstract

BackgroundThe metabolism of critically ill patients evolves dynamically over time. Post critical insult, levels of counter-regulatory hormones are significantly elevated, but decrease rapidly over the first 12–48 hours in the intensive care unit (ICU). These hormones have a direct physiological impact on insulin sensitivity (SI). Understanding the variability of SI is important for safely managing glycaemic levels and understanding the evolution of patient condition. The objective of this study is to assess the evolution of SI over the first two days of ICU stay, and using this data, propose a separate stochastic model to reduce the impact of SI variability during glycaemic control using the STAR glycaemic control protocol.MethodsThe value of SI was identified hourly for each patient using a validated physiological model. Variability of SI was then calculated as the hour-to-hour percentage change in SI. SI was examined using 6 hour blocks of SI to display trends while mitigating the effects of noise. To reduce the impact of SI variability on achieving glycaemic control a new stochastic model for the most variable period, 0–18 hours, was generated. Virtual simulations were conducted using an existing glycaemic control protocol (STAR) to investigate the clinical impact of using this separate stochastic model during this period of increased metabolic variability.ResultsFor the first 18 hours, over 80% of all SI values were less than 0.5 × 10-3 L/mU.min, compared to 65% for >18 hours. Using the new stochastic model for the first 18 hours of ICU stay reduced the number of hypoglycaemic measurements during virtual trials. For time spent below 4.4, 4.0, and 3.0 mmol/L absolute reductions of 1.1%, 0.8% and 0.1% were achieved, respectively. No severe hypoglycaemic events (BG < 2.2 mmol/L) occurred for either case.ConclusionsSI levels increase significantly, while variability decreases during the first 18 hours of a patients stay in ICU. Virtual trials, using a separate stochastic model for this period, demonstrated a reduction in variability and hypoglycaemia during the first 18 hours without adversely affecting the overall level of control. Thus, use of multiple models can reduce the impact of SI variability during model-based glycaemic control.

Glycemic control in the intensive care unit: A control systems perspective

“…Perhaps more importantly, such model-based SI [19,33,85,109,110,114,117,137,154,161,[165][166][167][168][169][170][171][172][173]] can be monitored and its level and/or variation assessed relative to condition [16,18,19,153,171,[174][175][176][177][178][179][180][181][182]. Thus, if accurate, this value offers not only the potential of good, personalised control, but also further insight into patient condition.…”

“…STAR is the only TTR system and controls both insulin and nutrition inputs [188,189], thus using both possible control inputs. It uses a unique risk-based stochastic forecasting [71] based on unique and increasingly complex stochastic models of future insulin sensitivity variability over 1-4 hours ahead [166,167,176,177,190,191]. The approach is used in both the ICU [156,159] and NICU [87,138,158,170], where notably insulin is the only control variable in the NICU for clinical reasons.…”

Stochastic Modelling of Insulin Sensitivity for out of Hospital Cardiac Arrest Patients Treated with Hypothermia

“…., S I ,0 ) to guide real-time clinical control (Evans, Shaw et al 2011;Fisk, Le Compte et al 2012). Thus, it is potentially important that the model is also as cohort-specific as possible for greatest accuracy and to minimize over-conservative forecasts (Thomas et al 2014).…”

“…Since stochastic modeling has shown its ability to quantify the probability of a future S I (Lin et al 2008), the resulting distribution of BG concentrations that would result from a given intervention can be determined (Lin et al 2008;Le Compte et al 2010;Evans et al 2011;Fisk et al 2012) . This information can be used to guide both insulin and/or nutrition interventions, which is the key to avoid unintended hypoglycaemia, improve overall glycaemic control, and identify periods of potential high glucose variability that may be indicative of unusual clinical events or cohorts (Thomas et al 2014) . This paper presents the adaptation of a stochastic model for S I prediction from adult critical care to the unique clinical and physiological case of OHCA patients, treated with hypothermia.…”

Stochastic Model Predictive (STOMP) glycaemic control for the intensive care unit: Development and virtual trial validation

“…Figure 2 indicates the relationship between the insulin sensitivity and the associated blood glucose trajectory. The model covers a broad medical ICU cohort over all the days of stay, but can be made specific to unique cohorts [39,40].…”