Today the Van de Bergh group in Belgium released a RCT investigating tight versus liberal glycemic control in the ICU. Before diving into this study, let's take a walk down memory lane.
Major interest in tight glycemic control in the ICU began in 2001 with an RCT by the same group of investigators, also published in the New England Journal. (11794168) Their initial study found that tight glycemic control (targeting a glucose of 80-110 mg/dL) caused a ginormous mortality benefit:
This study led to the creation of thousands of protocols to enact tight glycemic control throughout the world. All hail the insulin drip!
Subsequently, the NICE-SUGAR trial was a multi-center RCT that attempted to replicate the benefits of tight glycemic control. (19318384) Unfortunately, this study found that tight glycemic control actually increased mortality (likely due to an increase in dangerous hypoglycemic events):
Why did NICE-SUGAR show a mortality increase with tight glycemic control, whereas the 2001 Van de Bergh study found a mortality benefit from tight glycemic control? It's not entirely clear. But several theories emerged, including:
- Van de Bergh 2001 was a single-center study performed in a single surgical ICU. Sometimes single-center trials don't replicate very well, due to some foible of therapy at that specific ICU.
- Patients in the 2001 Van de Bergh trial were treated with early parenteral nutrition, which tends to cause a spike in glucose (unlike most other hospitals that don't utilize early parenteral nutrition).
- The Van de Bergh 2001 trial reported a lower rate of severe hypoglycemia in patients recieving aggressive glycemic control (5%) as compared to the NICE-SUGAR trial (7%). So maybe ICUs in Belgium are exceptionally good at titrating insulin safely (remember this bit – we'll get back to it later).
Overall, it certainly appeared that the NICE-SUGAR trial utilized superior methodology and obtained a more robust result. Thus, on the whole, institutions moved away from tight glycemic control. However, some doubts lingered.
new trial by Van de Bergh et al.: the TGC-Fast Trial
This fresh RCT by Van de Bergh should help erase any lingering doubts. This is a large multicenter RCT involving ICU patients randomized to two therapies: (37754283)
- Tight glycemic control (targeting 80-110 mg/dL).
- Liberal glycemic control: Insulin was only initiated when the blood glucose was >215 mg/dL on two consecutive measurements. Subsequently, insulin was utilized to target blood glucose levels between 180-215 mg/dL. Please note that this is a fairly liberal strategy – more liberal than is recommended by many current guidelines.
And the results demonstrated no difference in mortality! Womp womp.
What is going on here? One study shows benefit, the next shows harm, and the next shows no difference!
1) safety of tight glycemic control & hypoglycemic events
Perhaps the most important thing to notice here is that in the new 2023 study, there was an insanely low rate of severe hypoglycemia in patients receiving tight glycemic control (1% of patients developed glucose <40 mg/dL). This may be compared to 5% in the 2001 Van de Bergh trial and 7% in the NICE-SUGAR trial. Indeed, there was not a statistically higher rate of severe hypoglycemia among patients in the tight glycemic control group (47/4608 vs. 31/4622, p=0.07) – which is downright weird.
The Van de Bergh authors should be congratulated on achieving this bizarrely low rate of severe hypoglycemia (with the use of a computer algorithm to dictate insulin titration). However, this profoundly restricts the external validity of their trial. If most real-world hospitals attempted to institute tight glycemic control, their results would be dramatically worse. In reality (outside of the confines of a tightly regulated trial) the rates of severe hypoglycemia would probably be high (my guess is well above 7%).
Some folks might interpret this trial to mean that tight glycemic control is safe. In my view that a wholly incorrect interpretation (unless you happen to work in one of these magical 11 ICUs in Belgium).
2) efficacy of tight glycemic control
But what can we take away from this study? Although it is lacking in external validity (generalizability), it does seem to have robust internal validity (it's a well-done multicenter RCT without obvious sources of confounding).
This study compared very tight glycemic control (80-110 mg/dL) versus very liberal glycemic control (targeting <215 mg/dL) – and there were no differences in outcome. There were no differences in the primary efficacy nor the primary safety endpoints (ICU length of stay and mortality, respectively). The authors also looked at 23 secondary endpoints and a few of them showed some benefit – but this likely reflects statistical fishing rather than true benefit (if you correct for multiple statistical analyses, the results of these secondary endpoints lose statistical significance). Notably, none of the secondary endpoints related to infection showed any sign of benefit (despite reduced infection rate often being touted as a benefit of tight glycemic control).
Perhaps even more notable, there wasn't benefit in any patient subgroup (not even in cardiac surgery patients – a subgroup where folks are often enthusiastic about glycemic control):
My take-away from this analysis is that glycemic control itself doesn't affect outcomes. Despite implementing tight glycemic control with an unbelievable level of safety, this study failed to detect meaningful benefit in any subgroup of patients.
- The new TGC-FAST trial by Van de Bergh et al. achieved tight glycemic control with an incredibly low rate of hypoglycemia (1%) by using a computer algorithm to drive insulin delivery. Available evidence indicates that this level of safety cannot be replicated in real-world practice.
- In this magical Belgian world where tight glycemic control is bizarrely safe, glycemic control had no effect on clinical outcomes (even when comparing targets of 80-110 mg/dL versus <215 mg/dL).
- Overall, this study has substantial internal validity, demonstrating that glycemic control doesn't affect outcomes.
- If we attempt to extrapolate beyond an RCT context to actual practice in most hospitals (where tight glycemic control is more dangerous), this study may imply that liberal glycemic control is preferable.
Further discussion: IBCC chapter on glycemic control.
- PulmCrit: New ARDS guidelines reveal a shambolic state of affairs - December 10, 2023
- PulmCrit wee – Loading dose pharmacokinetics for antibiotics - December 8, 2023
- PulmCrit blogitorial: Why I don't believe the AMIKINHAL trial - October 26, 2023