29th International Symposium on Intensive Care and Emergency Medicine (ISICEM)

March 24-27, 2009; Brussels, Belgium — Day #1 Draft

The NICE-SUGAR Study Results


Prof. Simon Finfer, MBBS, FRCP, FRCA, FJFICM (Sydney, Australia)

The NICE-SUGAR study examined ~6,000 patients in over 35 ICU’s in Australia, New Zealand, Canada and the US. It is the largest study to date to look at the effects of tight glycemic control (TGC) in a critical care setting. NICE-SUGAR compares two glycemic control regimes – intensive insulin therapy (IIT) of 81-108 mg/dl versus conventional therapy (CT) of 144-180 mg/dl. The conclusions presented apply to adults expected to spend three days in the ICU, with predominantly enteral nutrition. (As a reminder, enteral means fed through a tube rather than by IV.)

The results of the trial were presented by Dr. Finfer, to a packed lecture theatre of at least 1,500 attendees – standing room only, literally, with people standing in the aisles. Our Q&A with Dr. Finfer, held after the presentation, is appended to this report.

The bottom line was that 90 day mortality was higher in the IIT group, and it was statistically significant (p-value of 0.02). At 90 days, the percentage of deaths in the IIT group was 27.5% vs. 24.9% in the CT group with an odds ratio of 1.14. Dr. Finfer stated, “We do not recommend targeting normoglycemia in critically ill adults.”

Differences in mortality of the two groups had emerged by day 28, although at that point they were not statistically significant. The cause of death was more likely to be cardiovascular in the IIT group vs the CT group. Apparently, sub-group analysis didn’t show any significant differences in mortality, but we will hear the full details tomorrow. Interestingly, there was no difference in secondary outcomes such as days in ICU, days in hospital, days on ventilator or dialysis.

Hypoglycemia was higher in the IIT group, as expected. In the IIT group, severe hypoglycemia (<40 mg/dl) was 6.8% vs. 0.5% in the CT group. It could be that this residual hypoglycemia made a difference in mortality, although at this stage we don’t know for sure and we imagine that would be difficult to determine, although Dr. Finfer said that they will perform an analysis in future. Equally, there may be criticism of the algorithm, but the overall results imply that it did its job well. Thus, the standard is likely not target a level as low as 81-108 mg/dL but rather a higher level may be more appropriate – some suggested <15o mg/dl. One question that has come up repeatedly in our discussions has been how “intensive” and “conventional” are defined – many US hospitals aren’t close to even the conventional range of 144-180 mg/dL, on average1. Another question is how well the units in NICE SUGAR adhered to protocol for checking for hypoglycemia.

Bottom line, it is interesting in our view that this is seen as “conventional” level in the trial when that level is still much lower than the average ICU in the US. It is still challenging enough to get many patients to this “conventional” level – as the higher average indicates, very few in the treated arm actually reached the lower level. More accurate blood glucose measuring equipment and CGM should have a role to play in this – even if the standard of care is <150mg/dl.

The NICE-SUGAR results echoed an earlier presentation at this conference, in which it was shown that clinical trials in ICU across many interventions have overwhelmingly yielded ‘negative’ results (although some were underpowered). Another general issue seems to be the heterogeneity of the patients and their conditions and how people reached their target numbers. It seems counter intuitive that high blood glucose is beneficial, and if so, the devil, as they say, will be in the details – which patients, how the protocol is applied, for how long, and how we avoid hypoglycemia. We still have a lot to learn.

  • The NICE-SUGAR (Normoglycaemia in Intensive Care Evaluation and Survival Using Glucose Algorithm Regulation) study is a 6,000 patient multi-center randomized controlled trial of blood glucose management comparing two insulin regimens for patients in critical care. The first regimen is intensive insulin therapy (IIT) to maintain blood glucose between 4.5-6.0 mmol/L (81-108mg/dL) versus conventional therapy (CT) - an insulin regimen to maintain blood glucose between 8.0-10.0 mmol/L (144-180 mg/dl) by administering insulin when blood glucose rose above 10.0 mmol/L. All patients were treated within the ICU for at least one full calendar day and had an arterial line placed as part of routine ICU management.
  • The primary outcome measure of the study is all-cause mortality within 90 days of admission. Secondary outcome measures include all- cause mortality at day 28 of the study, length of ICU stay, length of hospital stay, need for organ support (i.e. dialysis), incidence of blood stream infections, incidence and severity of hypoglycemia, and Extended Glasgow Outcome Score (Editor’s Note: this score includes the following eight categories — Dead, Vegetative State, Lower Severe Disability, Upper Severe Disability, Lower Moderate Disability, Upper Moderate Disability, Lower Good Recovery, and Upper Good Recovery).
  • The trial was established because of concerns in generalizing the Leuven results of 2001 from Dr. Greet Van Den Berghe’s group. Particular concerns included the fact that it was based mostly on parenteral nutrition (intravenous feeding), while the prevailing standard in Australia was enteral (via the digestive tract). Follow up trials, such as GLUCONTROL failed to replicate the beneficial effect. Additionally, in many of these glucose trials, hypoglycemia was very high – which may well confound the result. Hence the NICE-SUGAR trial set out to prove a ‘negative’ hypothesis, which was that there was no difference in relative risk of death between intensive and conventional therapy. Notably, NICE-SUGAR was very highly powered (90% power to detect a 3.8% difference in absolute mortality from an estimated baseline of 30%).
  • The baseline characteristics of the two groups were very similar – roughly 3,050 were randomized in each group, average age was 60, BMI 28 kg/m2, 20% had diabetes of which about 30% were already treated with insulin. Blood glucose at baseline was ~145 mg/dl. 93% of patients were on a ventilator.
  • In the trial, the average blood glucose in the IIT group was 115 mg/dl compared with 144 mg/dl in the CT group. Thus, very few people were likely at the 80-108 mg/dL baseline target for that group. In the IIT group, 97% of patients required insulin versus 69% in the CT group. The mean time on insulin was about four days in each group, but the amount of insulin per day was substantially higher in the IIT group (50 units daily vs. 17 units). Glucose control was obtained using a complex web-based algorithm.
  • Severe hypoglycemia (<40 mg/dl) was 6.8% in the IIT group vs 0.5% in the CT group. Nutrition was 80% enteral – quite different to the Leuven study. Also, steroid use was higher in the IIT group.
  • 90 day all-cause mortality was 27.5% in the IIT group vs 24.9% in the CT group, with an odds ratio of 1.14 and a p value of 0.02 (statistical significance is below 0.05). At 28 days, it was 22.3% vs. 20.8%, and this was not statistically significant. A Kaplan-Meier chart shows that differences in the two groups only start to emerge after about 20 days. A greater proportion of deaths in the IIT group was due to cardiovascular disease (42% vs 36%). Interestingly, there was no difference in secondary outcomes such as days in ICU, days in hospital, days on ventilator or dialysis.
  • It was stated that a sub-group analysis showed no major differences in the general result, although we await tomorrow’s presentation. Sub-groups considered included operative patients, people with diabetes, sepsis, trauma, APACHE II >25, and those on steroids.
  • The conclusions presented apply to adults expected to spend three days in the ICU, with predominantly enteral nutrition, irrespective of whether or not they have diabetes, or are operative. Note that the study draws no conclusions for those on parenteral nutrition. We are not sure what if any bias would have been introduced given this selection process, although Dr. Finfer believes that enteral nutrition is more prevalent across the world.
  • As we understand it, according to 2007 Medical Automations Systems (MAS) data, which collects glucose data directly from point of care devices, of all hospitals who indicated they were practicing tight glycemic control, the average in ICUs was 161 mg/dl – so we assume the averages in ICUs not practicing tight glycemic control are considerably higher. It is ironic that the average for those hospitals who say they are practicing TGC is squarely in the “conventional” range for NICE SUGAR. The highest average glucose of these ICUs was 220 mg/dL (again, for an ICU said to be practicing TGC) and the best hospital (NY‐based) had an average blood glucose of 128 mg/dl. By contrast, other units that are extremely focused have better results, such as Dr. Tony Furnary’s cardiac surgery ICU unit in Portland, which has an average 104 mg/dl in 2007 and very low hypoglycemia, unlike the group in NICE SUGAR.


Q&A with Dr. Simon Finfer

Dr. Finfer kindly spent some time talking to us after his presentation - our Q&A session with him is below. His conclusion was that the trial proved that the ’across the board’ approach taken by many hospitals of lowering glucose is not beneficial for all patients – but that that doesn’t mean we don’t have a lot more to learn about the topic. He is also concerned about the ability of ICUs to maintain tight control within a desired range – and in this respect he’s interested in more accurate blood glucose measurement and CGM. It occurs to us that for this reason, the closed loop might also be beneficial in the hospital.

Q: The conventional therapy group appeared very well controlled – their average blood glucose was 144 mg/dl - how was that?

A: Remember that 30% of the CT group never had insulin at all which means they never got higher than 180 mg/dl, and for the others, the algorithm was tuned to maintain the maximum blood glucose somewhere between 8 and 10 (144 mg/dl and 180 mg/dl). The patients also experienced lower blood glucose as they got better and insulin was discontinued.

Q: So will 144-180mg/dl be the standard of care?

A: This is the largest trial to date, and we have the most generalizable data, so we have to recommend that this is better than more intensive control.

Q: Does hypoglycemia cause mortality?

A: We have over a million hours of blood glucose data from our trial. We will analyse this data in a lot more detail, and it might well be that lesser degrees of hypoglycemia contribute to mortality, but we don’t know that yet. This analysis can be complex, but we will attempt to do it.

Q: How was blood glucose measured?

A: We used whatever the standard for each hospital was – typically point of care blood glucose meters. The other issue that is still unanswered (and important) is the measurement issue – because we used point of care measurement, which may not be accurate enough to target a tight range of blood glucose.

Q: Does insulin have any other effects?

A: There is evidence that it does indeed have many other effects, but at this point we can’t attribute the results of our trial to that – it’s too early to say.

Q: Is there any affect in the different sub-groups?

A: The treatment effect was the same for people with and without diabetes, for the less and more sick. The two groups where we can’t be absolutely confident about were the trauma patients and those receiving corticosteroids before hospitalization.

Q: The Kaplan-Meier mortality graph only starts to diverge at 20 days. Why is that?

A: Although the median treatment time was four days, there were many who stayed longer in the ICU. Two thirds of the deaths occurred in the ICU, but they were later, because we are very good at keeping people alive in the ICU for a long time. So one of the important results from this trial is that targeting 28 day mortality may miss the main result.

Q: How does this result fit with the Leuven study of 2001?

A: There are a number of differences between the trials – for example the feeding is different. One school of thought is that parenteral (intravenous) feeding leads to higher mortality in the control group because blood glucose can get too high. Feeding surveys conducted internationally suggests that enteral nutrition is much more prevalent around the world. In the Leuven trial, the low limit for the control group was also 218 mg/dl, which would now be too high for many ICUs. Also, methodologically we think that we did a very clean trial. I would expect that our result is much more likely to be generalizable.

Q: So are there specific patient groups that do need tight glycemic control – e.g. cardiovascular patients with pre-existing diabetes?

A: For these groups there is observational data only, and there are so many other variables that are constantly changing – so drawing conclusions is fraught with danger. So there is no randomized controlled trial data that would convince me to treat any groups of my patients differently.

Q: What is the role of glucose monitoring (CGM) in the ICU?

A: As I said before in JAMA, the next step is to do some Phase II multi-center efficacy studies, with a very accurate system of measurement and prove we can maintain tight control in the range we want. If we can show that, then we could move to a Phase III type trial with CGM – I think that is the way this field ought to move forward. This is a difficult treatment to deliver – it’s not like taking a pill!

Q: There were more deaths were from cardiovascular causes – why?

A: More patients in the intensive group received corticosteroids, and more insulin. Insulin may be causing problems before the cardiovascular system has stabilized. I’m sure you are aware of the ACCORD study, which also raised cardiovascular questions in type 2 diabetes.

Q: What might be the ultimate implications of this trial?

A: The international guidelines on glycemic control need to be revised. Given that intensive therapy consumes more resources, it’s not to be recommended. I do think that we need to do more research, particularly if we can use more accurate measurement systems.

My goal has been to improve the way that we do research. The number of participants in most previous ICU randomized controlled trials was too small to get positive results. More information can only benefit patients.

It’s early days yet. No one clinical trial has ever put an issue to bed. But we’ve answered one question – we shouldn’t continue doing what a lot of people around the world are doing – believing that if we just try and lower blood glucose we will be benefitting our patients. That doesn’t mean we don’t have a lot more to learn.


Our additional questions on NICE SUGAR are listed below:

  • How were the targets chosen? What are the implications of the results given that many centers in the US practicing TGC have averages well above the conventional blood glucose average reached?
  • How many centers were actually able to adhere to the glycemic control protocols? How frequently was blood glucose checked versus the protocol instructions?
  • How much would hypoglycemia be reduced with accurate, monitoring?
  • How does this result affect the thinking about whether TGC benefit is due to the effects of insulin?
  • Could blinding be better addressed if we had sensor augmented automated insulin infusion protocols suitable for use in the ICU?
  • What kind of bias was introduced by trying to predict who would have a length of stay greater than three days?
  • How much weight should be put on the fact that a substantial number of patients in the intensive arm were "modestly above target?" Will there be a subgroup analysis on outcomes for those that reached targets?
  • How does this result affect the criticisms of excessive parenteral feeding in Leuven I (and II)?
  • What were the nurse to patient ratios?
  • What is normoglycemia considered to be a critically ill population? What should it be?


-- John Close, Kaku Armah, Brendan Milliner, and Kelly Close