In our updated meta-analysis of randomized trials of intensive insulin therapy in critically ill patients, we found that such therapy had no effect on the overall risk of death. By including data from the largest trial of intensive insulin therapy, which was recently published,18
we provide the most current and precise estimate of the effect of intensive insulin therapy on mortality and severe hypoglycemia in the ICU setting. We found significant heterogeneity between studies, which was driven primarily by the 2 trials involving surgical patient populations.8,29
In keeping with this observation, our meta-regression analysis suggested that intensive insulin therapy may benefit patients in surgical ICUs. Finally, there was a 6-fold increased risk of severe hypoglycemia among patients given intensive insulin therapy compared with the control treatment. The risk of hypoglycemic events did not differ by type of ICU, or by intensity of insulin therapy.
Our meta-analysis showed a similar overall pooled estimate of effect on mortality, and similar confidence intervals, as the meta-analysis by Wiener and colleagues.14
One important difference between these 2 reviews is our finding that the effect of intensive insulin therapy differed by ICU setting, with a benefit demonstrated among surgical patients. Weiner and colleagues did not find this. These discordant results may be explained by the inclusion of different trials. We excluded 3 unpublished trials that had been included by Wiener and colleagues. Although we found no evidence of publication bias in our analysis, the tests we used may be unreliable in the presence of significant heterogeneity.41
However, since neither peer reviewers nor we have been able to assess the methodologic quality of the unpublished trials, we decided to include only trials published either in full or as abstracts in indexed journals.42
In addition, following translation, data abstraction and quality assessment, we included 1 trial that had been excluded by Wiener and colleagues;29
this trial was conducted in a surgical ICU and strongly favoured intensive insulin therapy. Although the NICE-SUGAR study included 2233 surgical patients and mortality was significantly increased in this subgroup, this study was conducted in mixed ICUs and was analyzed accordingly in our meta-analysis. The increased mortality in surgical patients enrolled in the NICE-SUGAR trial, the majority of whom were admitted to the ICU following emergency surgery, suggests that the benefit of intensive insulin therapy in patients treated in surgical ICUs requires confirmation.
Our findings of a significantly increased risk of hypoglycemia with intensive insulin therapy are in keeping with those reported by Wiener and colleagues.14
However, the analysis of differences in the risk of hypoglycemia between subgroups is limited by the low number of included studies.
Finally, our findings do not support the guidelines of organizations such as the American Diabetes Association, the American Association of Clinical Endocrinologists and other organizations, including the Surviving Sepsis Campaign, who recommend intensive insulin therapy for all critically ill patients.15–17,43
Our meta-analysis incorporates the results of the largest trial to date. We are not aware of any ongoing trial of sufficient size to affect these results; thus, we suggest that policy-makers reconsider recommendations promoting the use of intensive insulin therapy in all critically ill patients. However, ours was a pooled analysis of trials conducted in different populations with differing severity of illnesses and event rates; therefore, we cannot exclude the possibility that some patients may benefit from intensive insulin therapy. A meta-analysis of individual patient data and additional randomized trials may help to delineate which patients may benefit from intensive insulin therapy and be at lower risk of hypoglycemic events.
This review highlights a number of important questions that remain to be answered; most notable, why the beneficial treatment effect observed in Van den Berghe’s first trial8
has not been observed in subsequent multicentre trials involving adults. First, it may be that patients admitted to surgical ICUs after elective surgery benefit from intensive insulin therapy and that subsequent trials have not adequately examined this subgroup. Although this is suggested by our meta-regression showing a beneficial effect in patients admitted to surgical ICUs, subgroup analyses must be interpreted with caution, particularly in the setting of low event rates.44,45
In the surgical ICU subgroup, there were only 77 deaths in the intensive insulin therapy group and 110 in the usual care group.
A second possibility is that, because intensive insulin therapy is a complex treatment and the results may be dependent on the implementation of the intervention and the accuracy of blood glucose measurement,46
these factors may have differed between Van den Berghe’s first trial and subsequent trials. There is some evidence for this in the wide range of hypoglycemic events in the intervention arms between trials (from 5.1%–28.6%). A further explanation is the considerable variability in what constitutes “usual care.” In the trial by Van den Berghe and colleagues, the control group targeted a blood glucose level of 10.0–11.1 mmol/L. In contrast, some of the trials included in our analysis used a lower glucose target in the control group. For example, in the control arms of the trials by the NICE-SUGAR group and by Oksanen and colleagues,36
the targeted blood glucose level was less than 10.0 mmol/L and 6.0–8.0 mmol/L, respectively. In settings where usual care is to target a blood glucose level of less than 10.0 mmol/L, intensive insulin therapy may offer no benefit.
Another possible explanation for the discordant results between trials is that the degree with which blood glucose levels fluctuates in an individual patient may be as important as the average blood glucose concentration achieved.47
It is possible for 2 trials to report having achieved similar mean blood glucose concentration while obscuring the fact that blood glucose variability was markedly different between the 2 trials. Finally, the impact of feeding regimens on the effect of intensive insulin therapy requires urgent clarification. The patients in the trial by Van den Berghe and colleagues8
were fed largely by the parenteral route and received large doses of intravenous glucose.9
In contrast, patients in other studies, such as the NICE-SUGAR study, received predominantly enteral nutrition.18
We hypothesize that the treatment effect of intensive insulin therapy may be dependent on the means of controlling blood glucose, the accuracy of blood glucose monitoring, the degree of within-patient variation in blood glucose and the feeding regimen used. These hypotheses can be further explored through meta-analysis of individual patient data48
and by rigorous, adequately powered and carefully conducted randomized controlled trials.