The present study observed that implementation of IIT, with the percentage of patients receiving insulin by infusion increasing from 9% in period I to 42% in period III, was associated with no hospital mortality benefit. Furthermore, an incremental mortality increase associated with IIT was observed in patients with an ICU stay of 3 days or less. These latter findings are consistent with those from a similar subgroup in a randomized controlled trial (RCT) of IIT in medical ICU patients [24
Our study has several potential limitations. Despite our effort to control for confounding by assessing patient characteristics across the three study periods and adjusting for any baseline differences seen, the possibility of bias remains. However, arguing in support of a true incremental mortality associated with tighter glucose control was that the trend observed was opposite of what might have been expected if confounding had been present, given that several other clinical protocols designed to improve patient outcomes were implemented in the ICUs of the study hospital at around the same time as the IIT protocols. Admittedly, two of these protocols would not be expected to have any appreciable effect on mortality (invasive diagnosis of ventilator-associated pneumonia and ventilator weaning protocol) and the third (a protocol for lung protective ventilation) was widely practiced prior to formal protocol release. We did not observe any mortality differences between the first and second halves of each study period, arguing against an overall trend of increasing in mortality during the study (data not shown). These arguments suggest that our approach to evaluating the impact of the implementation of an IIT protocol on mortality is a valid one.
The major strength of the current study lies in the large cohort of patients included (>10,000) and the availability of extensive clinical data that allowed adjustment for severity of illness across time periods.
Four prior published studies have observed at least some reduction in mortality associated with IIT in critically ill patients. Van den Berghe and colleagues [23
] reported the results of a large randomized trial of IIT in patients admitted to a surgical ICU, approximately 60% of whom had undergone cardiac surgery. IIT (blood glucose range of 80 to 110 mg/dL) was associated with a reduction in mortality from 8.0% to 4.6% compared with conventionally treated patients (blood glucose range of 180 to 200 mg/dL) in addition to reductions in multiple morbidities [23
]. In a subsequent study, van den Berghe and colleagues [24
] randomly assigned patients admitted to a medical ICU to IIT or conventional blood glucose management. There was no clear mortality benefit for IIT in the intention-to-treat population, whether for ICU (24.2% IIT group versus 26.8% conventional treatment group) or hospital (37.3% IIT versus 40% conventional treatment group) mortality. However, IIT was associated with reduced in-hospital mortality in those patients who remained in the ICU for more than 3 days. In two studies that reported outcomes before and after implementation of an intensive glucose management protocol, mortality after implementation of the protocol was improved compared with prior to protocol implementation [21
]. However, neither study adjusted for severity of illness or other factors that may have changed over time.
There are several reasons that may explain why the results of our study differ from these previous investigations of IIT. The implementation of progressively more aggressive insulin therapy protocols at our institution resulted in a large change in practice: the use of IIT rose from 9.6% of patients in period I to 42% in period III. Although this resulted in a reduction in mean daily and mean morning glucose concentrations, with a difference of approximately 15 mg/dL between periods I and III, we were unable to consistently achieve blood glucose concentrations within the range of 80 to 110 mg/dL. It is possible that the benefits of IIT are not achieved unless glucose concentrations are lower than 110 mg/dL.
Failure to achieve the targeted glucose levels reflects the difficulty in application of clinical protocols to real-world practice, outside the rigid confines of RCTs, and has been recognized previously [35
]. Despite an explicit protocol combined with continuing educational efforts to alert physicians and nurses to the potential benefits of tight glycemic control, glucose levels (on average) remained approximately 20 mg/dL above the target range. This 'failure' likely occurred at several levels, although we are unable to discern specific causes within the limitations of our study design.
A major difference between our cohort and that included in the trials of van den Berghe and colleagues [23
] is the form of nutritional support used. In the studies of van den Berghe and colleagues, nutritional support was very aggressive: parenteral nutrition was administered early in the course of care and comprised the vast majority of non-protein calories during the first few days of ICU stay. This is in contrast to the practice at our institution, where parenteral nutrition generally is not instituted until 3 to 5 days after ICU admission. Parenteral nutrition reduces endogenous glucose production and promotes hyperglycemia in critical illness, effects which may be modulated by insulin administration [38
]. It is not clear how these effects would translate into increased benefit from IIT.
Finally, our study differs from the trials of van den Berghe and colleagues [23
] in that we examined a mixed population of critically ill patients, with approximately 60% of patients carrying a surgical diagnosis at admission, with a high representation of trauma patients (approximately 56% of the entire study cohort), including those with neurological injury. Our data suggest that, if anything, ITT was associated with an increased mortality among trauma patients, although the reasons for this are not clear. Previous retrospective studies have found an association between hyperglycemia and mortality in trauma patients [1
] and possibly a lower mortality temporally associated with the implementation of an IIT protocol and reduction in glucose variability [22
]. However, our study is the first to examine the effects of IIT on outcome in this population, applying rigorous adjustments for patients' baseline characteristics. Cardiac surgical services are not performed at out institution; therefore, our surgical population differs from that of the study of van den Berghe [23
], in which the majority of patients were recovering from cardiovascular surgery.
Several other studies also did not observe evidence of benefit of ITT. A recently published multicenter RCT in 537 patients with severe sepsis found no difference in mortality or organ failure in IIT versus conventional glucose management groups [45
]. The incidence of severe hypoglycemia was increased in patients randomly assigned to IIT (17.0% versus 4.1%). Another recent study investigated the effect of intraoperative IIT on the outcome of patients undergoing cardiac surgery [46
]. Patients were randomly assigned to IIT (glucose range of 80 to 100 mg/dL) or conventional treatment (glucose of less than 200 mg/dL). The study reported higher mortality and higher occurrence of strokes in the IIT group. A third recent RCT found no mortality benefit and a much higher incidence of hypoglycemia in a group of patients treated to maintain glucose less than 108 mg/dL for 48 hours after cardiac arrest compared with a group treated to maintain glucose less than 144 mg/dL [26
]. A prospective consecutive series of 818 patients admitted to a trauma ICU found no reduction in mortality or infectious complications in association with the implementation of a normoglycemic management protocol (glucose goal of 80 to 110 mg/dL) [47
]. Additionally, preliminary results from another recently completed randomized trial of IIT in ICU patients showed no significant mortality difference and an increase in the risk of hypoglycemia [27
The reason that IIT may result in increased mortality is unclear but may be related to a direct effect of insulin or to insulin-induced hypoglycemia. Two previous studies have found an association between ICU mortality and insulin administration [49
]; the mechanism by which insulin may confer harm is not clear but might be related to anabolic effects, similar to growth hormone [51
]. Two studies have examined the association between hypoglycemia and mortality in critically ill patients using case control methodology, with one finding no effect [54
] and the other suggesting an independent association between severe hypoglycemia and mortality [55
]. Further exploration of these areas is warranted.
It is unclear why we found increases in ICU mortality in the entire cohort and in some subgroups whereas there were only trends toward increased hospital mortality (Tables and ). For hospital mortality, the signal in the data could be attenuated due to the underlying noise of hospital deaths not related to glycemic control. It is also possible that ICU mortality reflects more glycemic control-related deaths than deaths from all causes. The association between period and ICU or hospital mortality was stronger in patients with a short ICU stay. This finding was also observed in the second randomized trial of van den Berghe and colleagues [24
]. It is also possible, though speculative, that IIT confers some longer-term survival benefit that offsets adverse effects seen during the ICU stay.