The present study reveals that the most relevant factors associated with blood glucose concentrations < 80 mg/dl or ≥ 150 mg/dl are age and application of noradrenaline, steroids or insulin. In addition, neurosurgery and severe sepsis/shock are variables strongly associated with blood glucose concentrations ≥ 150 mg/dl.
Patients who underwent neurosurgery and those in septic shock [1, 2], are frequently treated with noradrenaline and steroids. In separate analyses of explanatory variables (Table 2), severe sepsis/septic shock was associated with increased risk of blood glucose concentrations ≥ 150 mg/dl. Hyperglycemia, i. e. plasma glucose > 120 mg/dl in the absence of diabetes, is one of the broadened diagnostic criteria for sepsis in the 2003 definition. The frequencies of severe sepsis and septic shock were higher and mortality rates lower when the 2003 definitions (lower thresholds for the definition of organ dysfunctions ) were applied instead of the 1992 definitions although it was in the same population of critically ill surgical patients . Elderly neurosurgical patients requiring noradrenaline to increase intracranial perfusion pressure and steroids to reduce edema and swelling, as well as elderly patients with more severe organ dysfunctions and in septic shock who were treated with noradrenaline and steroids according to the "Surviving Sepsis Campaign" guidelines [1, 2] are at higher risk for blood glucose concentrations ≥ 150 mg/dl.
In a recent meta-analysis, there was no survival benefit regarding tight glucose control with intensive insulin therapy (IIT) in all analyzed 13,567 patients (OR 0.93, 95% CI 0.83 - 1.04), however, in the subset of surgical patients (OR 0.63, 95% CI 0.44 - 0.90)  there was a significant survival benefit. In the perioperative concept of IIT, the greatest reduction in mortality in surgical ICU patients with sepsis and receiving mechanical ventilation involved deaths due to multiple-organ failure . In these patients with severe sepsis it was speculated that the higher rate of severe hypoglycemia in the IIT group  was due to renal or liver dysfunction . In the present study, surgical patients with renal dysfunction assessed as extracorporeal renal replacement therapy were not associated with an increased risk of blood glucose concentrations < 80 mg/dl or ≥ 150 mg/dl (Table 2, 3). This is in contrast to medical-surgical ICU patients, where hemodialysis was a predictor of hypoglygemia  and renal insufficiency was an independent risk factor for severe hypoglycemia . In the present study, severity of liver dysfunction was assessed by the MELD score and of organ dysfunctions by the SOFA score. In separate analyses (Table 2), a MELD score > 9 and a SOFA score > 4 were associated with a higher risk of blood glucose concentrations < 80 mg/dl or ≥ 150 mg/dl, however, this association was not found by multiple logistic regression (Table 3). In medical-surgical ICU patients, female gender was one of the main predictors of hypoglycemia . The higher risk of blood glucose concentrations < 80 mg/dl with age (Tables 2, 3) and female gender (Table 2) in the present study are most probably due to an impaired counterregulatory response . The higher risk of hypoglycemia may be due to the lower counterregulatory hormone threshold for hypoglycemia in women than in men, i. e., especially adrenaline, noradrenaline and growth hormone response . The association of blood glucose concentrations < 80 mg/dl with administration of noradrenaline and steroids was surprising. In a rat model of continuous endotoxin infusion resembling some of the metabolic and cardiovascular constellations of human sepsis, a decreased number of hepatic plasma membrane alpha 1-adrenergic receptors were found . This might in part explain why a comparable subset of our patients might not have developed blood glucose concentrations ≥ 150 mg/dl but < 80 mg/dl in association with continuous noradrenaline infusion. On the other hand, hyperglycemia due to insulin resistance has been reported in human sepsis. Adipokine levels were extensively altered in patients with severe sepsis and shock . Adiponectin, the prototype of an anti-inflammatory and insulin-sensitizing adipocytokine, was diminished in these patients . Out of the insulin-resistance mediating factors, plasminogen activator inhibitor-1 (PAI-1), MCP-1, IL-6, IL-8, IL-10 and tumour necrosis factor (TNF)-α were significantly elevated in these patients. All significant changes were shifted in the same direction as in obese subjects and patients with type 2 diabetes. These results may help to explain insulin resistance in critically ill patients and patients with systemic inflammatory response syndrome. After diagnosis of severe sepsis and shock, serum adiponectin levels and hydrocortisone correlated positively with insulin demand, and noradrenaline demand negatively with male adiponectin levels . Septic shock causes a massive dilation of the peripheral vascular system, promoted by inflammatory cytokines and microbial toxins. Thus, the dosages of noradrenaline needed to provide a sufficient mean arterial pressure can be used to assess the degree of deterioration of the circulation (Sequential Organ Failure Assessment, SOFA score) for severity of organ dysfunction . Thus, an increase in noradrenaline demand may reflect a clinical situation with changes in adipokines resulting in insulin resistance. Taken together, these data in septic patients may explain why we observed a positive association of noradrenaline with blood glucose concentrations ≥ 150 mg/dl. On the other hand, adipokine panels resulting in overwhelming insulin sensitivity might set patients at risk of blood glucose concentrations < 80 mg/dl. In this constellation, we might get an association of steroid and noradrenaline treatment with blood glucose concentrations < 80 mg/dl, which is not due to the effects of steroids or noradrenaline per se. The variables SAPS II > 36, MELD > 9, and SOFA > 4 in the present study were associated with an increased risk of blood glucose concentrations < 80 mg/dl reflecting patients with greater severity of disease and organ dysfunctions. One third of our patients were neurosurgical patients. Neurosurgical patients are often treated with noradrenaline to increase intracranial perfusion pressure and steroids to reduce edema and swelling. Thus, the pathopysiological reasons and the underlying adipokine pattern during steroid or noradrenaline treatment in our heterogenous patient groups may differ profoundly. This might explain why we found noradrenaline and steroid treatment as risk factors for both, blood glucose concentrations ≥ 150 mg/dl and < 80 mg/dl.
The blunted ability of neutrophils to adapt to physiological hyperinsulinemia in older people (69 +/- 4 years) may compromise the anti-infective response , and may contribute to sepsis. Taken together, elderly female patients with extracorporeal renal replacement therapy and/or liver dysfunction, especially under IIT, may be at increased risk of hypoglycemia.
The "Surviving Sepsis Campaign" guidelines for the management of severe sepsis and septic shock recommended intensive insulin therapy (IIT) targeting of glucose levels to the < 150 mg/dl range [1, 2]. In the present study, 48.4% of minimal and maximal blood glucose data sets per day were in the target range of 80 to < 150 mg/dl (Table 1). This is comparable with a previous study in a mixed medical-surgical ICU with strict glycemic control, in which less than 50% of patients were within the target range with a considerable overlap between the intensive and the standard insulin group . In the present study, severity of disease was assessed by SAPS II score. In separate analysis only (Table 2), SAPS II was associated with increased risk for blood glucose concentrations < 80 mg/dl or ≥ 150 mg/dl, however, not in multiple logistic regression (Table 3). Severity of illness reflected by the APACHE II score was one of the factors associated with the development of severe hypoglycemia in medical, surgical, and cardiac ICU patients in multiple logistic regression [10, 14]. Thus, the severity of disease may put patients at risk of not falling within the recommended target range of 80 to 150 mg/dl blood glucose concentration.
In the present study, the overall hypoglycemia rates observed with blood glucose concentrations < 40 mg/dl in 0.9% and < 80 mg/dl in 10.9% (Table 1) were low, furthermore the administration of insulin was associated with an increased risk of blood glucose concentrations < 80 mg/dl (OR 2.1; 95% CI 1.7 - 2.6) (Table 3). A comparison with hypoglycemia rates in the literature is difficult since hypoglycemia rates with IIT might be related to differences in patient populations, severity of illness, protocols, blood glucose targets, definitions of hypoglycemia, proportion of diabetic patients, ICU length of stay or duration of therapy. For example, the rates of hypoglycemia (≤ 40 mg/dl) were 0.5% , 1.7%  and 2.7%  in the control group, and 6.8% , 8.5% , 8.7%  and 16%  in the IIT group, respectively, in mixed surgical/medical ICUs. Taken together, in a recent meta-analysis, IIT was associated with a high risk (OR 6.0, 95% CI 4.5 - 8.5) for hypoglycemia < 40 mg/dl .
A comparison to patients with blood glucose levels always within the target range of 80 to 150 mg/dl, patients with episodes of blood glucose concentrations < 80 mg/dl, of ≥ 150 mg/dl, or of both, revealed an increased risk of death (Table 4). It has been suggested that quadrupling the rate of severe hypoglycemia and doubling the mortality attributable to severe hypoglycemia would erase the survival benefit of IIT in medical, surgical, and cardiac patients . It remains unclear, whether hypoglycemia is associated with worse outcome or is simply a marker of severity of illness . Spontaneous hypoglycemia without insulin therapy might be even more dangerous, resulting in a 1.7-fold increase in mortality , because it is often detected late, furthermore, the duration of the episodes are longer due to less frequent monitoring than with IIT.
Thus, it is of high clinical relevance to know that beyond insulin therapy, the most relevant factors associated with hypoglycemia were noradrenaline infusion, steroids and age (Table 3). To prevent deleterious hypoglycemia, a higher target range of 140 - 180 instead of < 150 mg/dl in critically ill patients was recently suggested [35, 36].
The strengths of our study are that our ICU has a high nursing staff to patient ratio and is operated 24/7 by an on-site Critical Care Board-certified intensivist, and that intervention-related risk factors for blood glucose concentrations ≥ 150 mg/dl were evaluated. On the other hand, our study has several limitations. It is a single center trial. Other factors include adherence to the intensive insulin therapy protocol, the frequency of glucose measurements, the amount of average calories/day, the insulin dose, and the small number of hypoglycemic episodes < 40 mg/dl, however most of the aforementioned limitations apply to many of the recently published studies concerning IIT as well.