Insulin sensitivity variability
Both cohort and per-patient results suggest that critically ill patients have significantly lower and more variable insulin sensitivity on day 1 than later in their ICU stay. Further analysis shows that this day 1 result is primarily influenced by the first 12–18 hours of ICU stay. Over this time, rapid improvements in insulin sensitivity level and variability occur so that there is no statistically significant difference between 18–24 hours and day 2. From day 2 onwards, changes in SI level and variability are not as large and of limited clinical and statistical significance.
Within the analyses, there are some differences in significance between cohort and per-patient results for comparisons after day 2. The overall findings noted in the preceding paragraph are the only clear, consistent trends across both analyses.
The counter-regulatory hormones: cortisol, glucagon, the catecholamines, as well as growth hormone are significantly elevated almost immediately after critical-insult, but decline rapidly over the first 12–48 hours [21–24]. These hormones are known to cause increased hepatic glucose production, inhibition of insulin release, and peripheral insulin resistance , all of which cause a decrease in the model-based SI metric used in this study. Hence, the low but rapidly increasing insulin sensitivity seen during the first 12–18 hours of ICU stay is likely due to the acute counter-regulatory response to critical illness.
Time in this study was referenced from the commencement of SPRINT, rather than ICU admission. However, the difference between admission time and commencing SPRINT was generally very short, with a median delay for this cohort of 1.9 hours. Within 6 hours of admission, 81% of the cohort had commenced SPRINT. Hence, these results are applicable to the first few hours and days of ICU stay.
The insulin sensitivity parameter
The model-based parameter used in this study represents a whole-body insulin sensitivity capturing overall metabolic response to exogenous insulin. SI captures the relative net effect of altered hepatic glucose production, peripheral and hepatic insulin-mediated glucose uptake, and endogenous insulin secretion. All of these effects are altered significantly in critical illness due to the stress response [25–27]. Hence, the metabolic balance that this parameter represents is an important consideration in TGC, because it determines a body’s glycemic response to exogenous insulin and nutrition.
As an identified parameter, SI contains unmodeled physiological effects and measurement device noise. However, Lotz et al.  indicated that this form of insulin sensitivity correlated very well (r > 0.9) with the “gold standard” euglycemic clamp and its change in a lifestyle intervention study on 73 normoglycemic healthy and obese subjects (146 clamp procedures before/after intervention). In the critical care setting, a similar version of the model and SI parameter has been cross-validated against independent, matched patient data from a single center of the Glucontrol randomized, clinical trial .
The analytic inaccuracy of bedside glucometers or any other sensor used to gather BG measurements influence individual values of SI. However, this study examines distributions of SI consisting of thousands of values identified from a wide range of BG values, thus both the random and bias components of error cancel out within each distribution. This effect was confirmed by Monte Carlo analysis (results not shown) using an error model for the glucometer derived from data supplied by the manufacturer .
Implications for tight glycemic control
With low and variable insulin sensitivity, glycemic levels may appear unresponsive and/or difficult to control effectively with exogenous insulin. This situation may provoke larger insulin doses from many protocols that have no explicit upper limits on insulin dose [6, 30–32]. High levels of circulating insulin coupled with the observed variability in insulin sensitivity result in increased glycemic variability and an increased risk of hypoglycemia during the first 24 hours of ICU stay.
Not only does glycemic variability pose a risk through hypoglycemia, it also is detrimental in its own right. Several studies [9–11, 33] have shown that glycemic variability is independently associated with mortality in critically ill patients. More specifically, Bagshaw  showed that hypoglycemia and variability within the first 24 hours of ICU stay are each associated with increased mortality. In vitro, high glycemic variability was shown to increase oxidative stress  and apoptosis , thereby suggesting a rationale to explain the clinical association with poor outcome.
Evidence from other studies [10, 12] indicates an association between hypoglycemia, glycemic variability, and mortality. However, the question remains: Is low and variable glycemia the cause of increased morbidity and mortality? Or is it just a symptom in very ill patients? Until this question can be answered conclusively, it is perhaps best to formulate TGC protocols not to exacerbate the situation, which requires the ability to differentiate more and less metabolically variable patients.
Another significant finding in this study is the range of variability seen across patients, as well as over time (Figures 3 and 5). Less variable patients, if identified, may be treated more aggressively with insulin without compromising glycemic variability. Hence, model-based methods have been mooted as a means of better managing this inter- and intra-patient variability [30, 36].
Only patients on the SPRINT TGC protocol were considered for this analysis as they had sufficient data density to identify SI hourly. Patients were put on the SPRINT protocol because they were hyperglycemic and thus were likely to be biased towards lower insulin sensitivity compared with other ICU patients. However, in the context of investigating the implications of SI variability on TGC, this cohort is appropriate.
Another limitation is the use of a model-based insulin sensitivity parameter, as it is not measured directly and may be influenced by modelling errors or un-modelled effects. As an identified parameter, SI contains unmodeled physiological effects and measurement device noise. However, as noted previously, this form of SI has been shown to correlate very well with the “gold standard” euglycemic clamp [17, 37] and has been shown to be an independent marker of metabolic condition . Finally, this method of analysis is robust to BG sensor error.
A further limitation is the relatively small cohort size available for analysis. The demands of manually transcribing written clinical data into electronic form and the specific inclusion criteria have restricted the number of patients for whom complete glycemic control data are currently available for analysis. The size of this cohort has precluded subgroup analyses, such as diabetic and cardiovascular surgery patients, because these subgroups only contain 20–40 patients. With relatively few patients, the subgroup analyses fail to demonstrate statistical significance, despite effect sizes and trends very similar to that seen in this overall analysis. Thus, these comparisons will be completed in the future, when more patient data become available.
The findings of this study should be equally valid in other ICUs where attention to TGC and blood glucose measurement frequency may be a lower priority. Although the data density might not be present to allow such units to explicitly identify SI hourly, these results indicate that patients will still have lower and more variable insulin sensitivity on day 1 than later in their ICU stay. Thus, suggestions of higher glycemic targets, conservative insulin dosing, and modulation of carbohydrate nutrition are especially pertinent.
Without the ability to identify patient-specific metabolic states, a protocol should be less aggressive over the first few days, and particularly the first 24 hours, to minimize variability. It may be important for protocols to consider higher glycemic targets on the first days of ICU stay (compared with later days) to ensure safety. Perhaps a glycemic target similar to the current guidelines of 7.8-11 mmol/L [38–40] is most appropriate for the first 24 hours with the target range, reducing over days 2 and 3 to more normoglycemic levels as SI level and variability improve.
Greater blood glucose measurement frequency and conservative insulin dosing can mitigate the impact of SI variability on risk  and also should be considered for the first few days of stay. Modulation of carbohydrate nutrition, within limits , can reduce the need for exogenous insulin to better manage glycemia .