Key findings
We studied clinical practice before and after a key multicentre randomised controlled study of glucose control in >170,000 patients from 49 ICUs in ANZ. Our findings confirm previously published survey data that IIT was not practiced in ANZ prior to NICE-SUGAR. Glycaemic control in all patients before NICE-SUGAR corresponded to the control group of the trial. We found that only minute changes in glycaemic control toward looser control followed the publication of the study. The change in glycaemic control applied similarly to all patient subgroups except for IDDM patients. The rate of severe hypoglycaemia remained unchanged, whereas the rate of moderate hypoglycaemia decreased by 14% after NICE-SUGAR. IDDM patients had highest Glu1−values, highest increase in Glu1 values, and highest rates of moderate and severe hypoglycaemia before and after NICE-SUGAR. The adjusted mortality decreased over the whole study period.
Relationship to previous studies
The mean Glu
1 for all patients in our study before NICE-SUGAR was 7.96 mmol/L. This is in accordance with the time-averaged mean glucose value of 8.0 mmol/L in the control group of NICE-SUGAR [
8]. In each subgroup, glycaemic control approximated to 8 mmol/L, except for patients with IDDM. As the number of IDDM patients represented only 3% of all patients before NICE-SUGAR, their impact on overall glycaemic control was small.
The quality requirement for data supply for individual ICUs was stringent. As the possible bias caused by missing data is of minor concern, it is reasonable to argue that the results of glycaemia in ICUs reflect the true practice of ANZ ICUs at that time. According to our results, IIT was not standard practice in ANZ as none of the ICUs had their median Glu
1 below the upper limit of the 95% confidence interval of glycaemia reported in the IIT arm of NICE-SUGAR [
8]. In an observational study conducted in ANZ ICUs and published in 2003, 41% of ICUs reported having implemented tight glucose control in their clinical practice for at least some patient subgroups [
17]. Simultaneous observational data, however, displayed a mean first day glucose value of 8.7 mmol/L [
23] suggesting that most ICUs in ANZ did not, in fact, practice IIT.
A nationwide translational study in Germany showed that 65% of ICUs perceived having tight glucose control in practice for patients with sepsis, but only 6% of patients were observed to be in the tight glycaemic range [
24]. There are no previous reports of national level translation of glycaemic control recommendations into practice in critically ill patients. Previous studies have shown that the translation of study knowledge of simple, cheap and well-recognized therapies may be poor [
25‐
27].
The loosened glycaemic control after NICE-SUGAR was reflected by the rates of moderate hypoglycaemia. The rates of moderate hypoglycaemia were roughly one-third lower than the 15.8% seen in the control arm of NICE-SUGAR [
10]. A partial explanation for this difference may be that we had only first day glucose measurements available. The median time of occurrence of moderate hypoglycaemia in NICE-SUGAR was one day, indicating that half of moderate hypoglycaemia occurred after the first day [
10]. There is no validation yet that the incidence of hypoglycaemia on Day 1 is a robust surrogate for the overall incidence of hypoglycaemia.
The rate of severe hypoglycaemia remained unchanged in the overall patient population as well as in the subgroups despite looser glycaemic control after NICE-SUGAR. The rates of severe hypoglycaemia in the control arms of the three RCTs were 0.8% [
6], 3.1% [
7] and 0.5% [
8]. The rate in our unselected patient population approximated that of the first Leuven and NICE-SUGAR studies. In medical patients, our clinical data showed much lower incidence of severe hypoglycaemia than the rate in the second Leuven study which included medical patients only. However, once again, our data are limited to Day 1 after ICU admission.
The patients with IDDM had higher Glu
1 than patients in any other subgroup, both before and after NICE-SUGAR. In the second Leuven or NICE-SUGAR studies, the mortality in diabetic patients was not different between the treatment groups [
7,
8]. In a large observational study, hyperglycaemia was associated with increased mortality in non-diabetic patients whereas in diabetic patients the risk was much lower, suggesting that patients with IDDM may have a different biological response to hyperglycaemia [
22] or that in patients with DM hyperglycaemia is indicative of a lesser degree of stress of illness severity. Interestingly, the Glu
1 values in our IDDM patients’ values approximated those in the above observational study suggesting that there might be greater difficulty in controlling glycaemia in diabetic patients or that tight glycaemic control is less vigorously pursued or both [
22].
IDDM patients had the highest Glu
1 of all patient subgroups along with the highest rate of both severe and moderate hypoglycaemia, reflecting the highest variability in glucose values of all patient subgroups. Although hyperglycaemia does not seem to be as harmful in diabetics as in non-diabetic patients [
22], higher variability of blood glucose values has been associated with increased risk of mortality in overall critically ill patients [
28]. Interestingly, we could not demonstrate a decrease in mortality in IDDM patients. Finally, overall mortality decreased during the study period, even when adjusted for disease severity. As the change towards looser glycaemic control in the overall population was clinically insignificant, it is unlikely to be an explanation for reduced mortality.
Implications of study findings
The glycaemic control before NICE-SUGAR in ANZ represented clinical practice at that time. It closely followed the glycaemic control achieved in the control arm of NICE-SUGAR enforcing the generalizability of NICE-SUGAR study results to ANZ and justifying the choice of the specific range of glycaemic level apply to the control arm of NICE-SUGAR.
Glycaemic control before NICE-SUGAR was not in the range of IIT in any of the ANZ ICUs included in our study. This is strong evidence that before the NICE-SUGAR study, ICUs in ANZ were not practicing glucose control according to recommendations at the time or failed to achieve targeted glucose control in their patients. The ICUs participating NICE-SUGAR loosened their glucose control practice after the publication of the study results to greater degree than the ICUs not participating in the trial. Overall, the translation of study results into practice seemed to be more prominent in academic ICUs as well as ICUs participating in the trial. This suggests that engaging ICUs in clinical research could facilitate translation of research into practice.
The extent of change in glycaemic control was clinically insignificant. It was, however, reflected by the lower incidence of moderate hypoglycaemia. As hypoglycaemia have been strongly associated with increased mortality, this change appears desirable.
Patients with IDDM were outliers in glycaemic control and hypoglycaemia rate. In addition, IDDM patients were the only subgroup in whom the mortality did not decrease over the study period. The recommendations for glycaemic control in critically ill patients do not separately address glycaemic control in IDDM patients. As this subgroup of patients seems to be highly different from the overall critically ill patient population as well as other patient subgroups, a different approach to their glycaemic control may be needed. If further studies on glycaemic control in critically ill patients are to be conducted, appropriate identification and stratified randomisation for this patient subgroup would be needed.
The observational period of our study was nearly five years leading to a large sample size of more than 170,000 patients. Accordingly, even for small changes in the numerical values between the study periods, statistical significance was easily achieved. In order to increase the robustness of our findings, a reduced P-value of 0.01 was chosen to indicate statistical significance. However, even with this conservative statistical approach, the clinical importance of some of the statistically significant changes seen remains unclear.
Strengths and weaknesses
Our study has several strengths. The data in the APD are prospectively collected for routine quality control purposes and, therefore, they are unlikely to be biased. We ensured the representativeness of the glycaemic control for each ICU by selecting only the ICUs that provided >80% of the first day glucose values for their patients. Some limitations should be noted, however. The completeness of first day glucose values for ICUs was low in the data collected 12 years ago. Thus, we could not analyse the effect of the first Leuven trial on clinical practice. It can be postulated that the greatest changes in clinical practice would have been before and after the publication of the first pivotal RCT with tight glucose control. However, by the time NICE-SUGAR was conducted, such changes (if they had occurred) had clearly dissipated. Secondly, we did not have glucose measurements beyond the first 24 hours of the ICU stay in the database. The first day mean glucose values, however, have been shown to have high accuracy in predicting the glucose control of the whole ICU stay [
19]. Third, our assessment of hypoglycaemia using Day 1 data is not validated. However, although the absolute overall incidence of hypoglycaemia may be inaccurate, the change from one period to another using the same assessment methodology is likely to represent a real change in its incidence within ANZ.